ICAME 2001 Future Challenges in Corpus Linguistics, Louvain-la-Neuve, Belgium, 16-20 May 2001
Paper Presentations and Work-in-Progress Reports
Bas Aarts, Evelien Keizer, Mariangela Spinillo & Sean Wallis
Hajar Abdul Rahim & Harshita Aini Haroon
Mia Bostrom Aronsson
Sylvie De Cock
Natalia Gvishiani & Oksana Gerwe
Hitoshi Isahara, Toyomi Saiga & Emi Izumi
Randall L. Jones
Fanny Meunier & Inge de Mönnink
JoAnne Neff, Emma Dafouz, Honesto Herrera, Francisco Martínez,
Juan Pedro Rica, Mercedes Díez, Rosa Prieto & Carmen Sancho
Using corpora to construct dictionaries for information retrieval and text
Susan Pintzuk & Ann Taylor
Robert Sigley & Janet Holmes
Nicholas Smith & Geoffrey Leech
Patrick Studer & Peter Schneider
Irma Taavitsainen, Päivi Pahta & Martti Mäkinen
Elena Tognini Bonelli
Gunnel Tottie & Hans Martin Lehmann
Joe Trotta & Mats Johansson
Niek Brom, Inge de Mönnink & Nelleke Oostdijk
Pam Peters & Adam Smith
Ann Taylor, Anthony Warner, Susan Pintzuk & Frank Beths
Maria Teresa Prat Zagrebelsky
Estelle Dagneaux, Sylviane Granger, Fanny Meunier, Stephanie
Petch-Tyson & Xavier Vilret
Martin Wynne & Oliver Mason
Charles Fillmore (University of California at Berkeley / International Computer Science Institute, USA)
A decade ago I offered some pronouncements on the opposition between reliance on corpus data for discovering and supporting linguistic generalizations, on the one hand, and the need to appeal to the intuitive knowledge of native speakers on the other hand1. In that context I had in mind people who were quite sure that only corpus evidence, or only introspection, was valid for doing proper empirical linguistics. My peace-making position was that one couldn't succeed in the language business without using both resources: any corpus offers riches that introspecting linguists will never come upon if left to their meditations; and at the same time, every native speaker has solid knowledge about facets of their language that no amount of corpus evidence, taken by itself, could support or contradict.
In the meantime I have been forced to face these same issues in the course of a six-year research project dedicated to learning facts about the English lexicon; in this project, corpus evidence is the main tool, and researchers who are speakers of English are hired to work with this resource and decide how to use what it gives us. Since for our purely lexicographic purposes, corpus evidence and our ability to interpret it provide more lexically specific information than can be found in dictionaries or lexical descriptions known to us, we are daily rewarded with insights about our language that introspection alone, however disciplined, could never direct us to. The limitation to lexical observations, of course, allows us to escape larger-scale and "deeper" kinds of linguistic facts: our work can proceed with "canonical" examples of the uses of the lexical units we target for study.
In this paper I will describe a lexicon-building effort called FrameNet and I will discuss some of the tensions my colleagues and I face between (a) the need for labor-intensive (and slow) work by linguistically sensitive trained annotators, and (b) the desire to maximize the use of computational means for organizing the material, facilitating the coding process, assigning tentative annotations, enhancing the editing and correcting activities, and collecting and summarizing results from the annotations. The project is dedicated to creating a "frame-based" lexicon cum thesaurus of modern English in which each lexical unit is described in terms of the semantic frame which underlies its interpretation and is provided with descriptions of its semantic and syntactic combinatorial properties, as attested by examples taken from the British National Corpus. I will sketch the most important aspects of the procedure and will characterize the database of annotated sentences as well as the means of deriving the combinatorial descriptions automatically from the annotations.
My discussion will cover the evolution of the project's goals from a fairly simply-defined starting place that emphasized governors and the nature of their dependents to a system that should allow us to provide useful information about implicit arguments, support verbs, selectionally transparent nouns (names of parts, types, aggregates, and quantities), frame inheritance and frame blending (where semantic frames are seen as complexes or blends of other frames), and more.
Christopher Tribble (King's College, London University, United Kingdom)
When we look at a corpus in order to find answers to questions, what is the nature of the thing we are observing, and what do we focus on? To date, the development of corpus linguistics has been largely motivated by the interests of descriptive linguists, lexicographers and grammarians, and literary scholars, along with language engineers and the natural language processing community. Although language teachers have indirectly benefited from the fruits of the labours of these different groups through the publication of new dictionaries and grammars, there has, as yet, been relatively little direct use of corpus data by language teachers and students in real language classrooms (despite earlier, highly sanguine predictions by distinguished corpus linguists - e.g. Sinclair 1991). In this paper, I shall outline some of the reasons why corpus data might, or might not, be being used in language teaching, summarise what has been achieved in this area despite possible problems (notably through the work of Tim Johns at Birmingham University), discuss why so relatively little use of corpora has been made in language classrooms to date, and outline ways in which corpora might make a greater contribution to language teaching in the future.
In developing this argument, I shall first have to assess the extent to which corpus data is an appropriate resource for language learners. In so doing I will re-visit a significant moment in the development of the relationship between language teaching and corpus linguistics (an exchange between H.G. Widdowson and J.M. Sinclair in the early 1990s (Widdowson 1991, Sinclair 1991, 1997). I shall also have to consider later arguments around which model (if any) of the English language is most appropriate for bilingual students (Amon 2000, Granger 1998, Pennycook 1999, Phillipson 2000, Seidlhofer 2000 & forthcoming, Jenkins forthcoming) and the implications that decisions in this area have for corpus design for language learners. This move is essential as I strongly hold that an uncritical adoption of corpus data per se is an insufficient stance when considering the needs of students and teachers of foreign or second languages.
Having established a position in relation to whether corpus linguistics might have something to say to language learners and teachers, and what corpus resources language learners and teachers might require to achieve their ends, I shall then demonstrate some of the areas in which I feel corpus linguistics can make a contribution to language learning and teaching. In this part of my presentation, I shall take as examples recent work by Ken Hyland (Hyland, 2000), Joanna Channell (Channell 2000), and some of my own work with Paul Thompson (Thompson, P. and C. Tribble forthcoming) and develop a case for the use of corpus resources and tools in general and special purposes language teaching - a case which asks corpus linguists to adjust their gaze if they wish to take into account the needs of the millions of students who are currently studying English.
In the last section of the paper, I shall present results from a recent email survey which assesses the extent to which language teachers do and do not use corpora in their teaching, and gives some insights into the reasons for what I consider to be a surprising under use of a valuable resource. I shall then propose some principles which teachers and students might use when considering using corpora in their own endeavours.
Ammon, U. (2000) Towards more fairness in international English: linguistic rights of non-native speakers? In Phillipson, R. (ed.) Rights in language. Lawrence and Erlbaum, London.
Channell J. (2000) Corpus based analysis of evaluative lexis. In Hunston, S. and Thompson, G. (eds) Evaluation in text. Oxford University Press, Oxford.
Granger, S. (ed.) (1998) Learner English on Computer. Longman, Harlow.
Hyland, K. (2000) Disciplinary discourses: social interactions in academic writing. Longman, Harlow.
Jenkins, J. (forthcoming) A sociolinguistically-based, empirically researched pronunciation syllabus for English as an international language, Applied Linguistics.
Pennycook, A. (1999) Pedagogical implications of different frameworks for understanding the global spread of English. In Grutzmann, C. (ed.) Teaching and learning English as a global language. Staufenburg Verlag, Tübingen.
Phillipson, R. (ed) (2000) Integrative comment: living with vision and commitment. In Phillipson, R. (ed.) Rights in Language. Lawrence and Erlbaum: London.
Seidlhofer, B. (2000) Mind the gap, English as a mother tongue vs. English as a lingua Franca, Views 9(1): 51-68.
Seidlhofer, B. (forthcoming) Closing a conceptual gap: the case for a description and pedagogy of English as a lingua franca, Applied Linguistics.
Sinclair, J.M. (1997) Corpus Evidence in Language Description. In Wichmann, A., Fligelstone, S., McEnery, T. and Knowles, G. (eds) Teaching and language corpora.: Longman, London and New York, pp. 27-39.
Sinclair, J.M. (1991) Shared knowledge. In Alatis, J.E. (ed.) Linguistics and language pedagogy: the state of the art. Georgetown University Press, Washington D.C., pp. 489-500.
Thompson, P. and Tribble, C. (forthcoming) Looking at citations: using corpora in English for academic purposes. In Tribble, C. and Barlow, M. (eds) Special issue on corpora in language learning and teaching, Language Learning and Technology 5(3) http://llt.msu.edu/.
Widdowson, H.G. (1991) The description and prescription of language. In Alatis, J.E. (ed.) Georgetown University round table on languages and linguistics 1991. Georgetown University Press, Washington, D.C., pp. 11-24.
So far has Natural Language Processing (NLP) moved in the last ten years that the title of this talk now has a strong semantic redundancy, far removed from the days when (according to a recent posting) Chomsky could say that there was no such thing as corpus linguistics, a sentiment that many in NLP might then have echoed, even if they had agreed with Chomsky on virtually nothing else! I grew up within the Artificial Intelligence (AI)/NLP tradition as a passionate anti-Chomskyan who was happy to satirise Chomsky’s remarks about the irrelevance of data but did not do a great deal of data gathering myself. However, that was not for any reasons of principle but rather that data was hard to store on primitive machines and, even when it was available, there was insufficient processing power for interesting computations. A classic example would be Sparck Jones’ (1966) thesis on Semantic Classification, which reclustered the data of Roget’s Thesaurus, but which actually contained no computational results because the matrices required by the clumping theory she used could not be fully computed in those days. Nonetheless, the thesis was, rightly, highly influential. My own thesis, contemporary with hers and from the same research laboratory, computed only over ten small philosophical texts and ten randomly chosen newspaper editorials as controls; an absurdly small sample by today’s standards but near the limit of what was feasible at the time.
Within that historical anecdote is a contrast of kinds of corpus, which will be important later in the talk: between prose corpora (what is normally meant by the word corpus) and corpora that are dictionaries or thesauri. All are man-made and consisting only of words, but the latter are in some sense metacorpora, and one looks in them not for usage but for facts about usage, based on observation and intuition.
The talk begins with a listing of NLP modules, or separable tasks, that are responsive to some form of learning over corpora (in the standard sense above), and which use supervised or unsupervised methods or both. Some classic corpora are named, and a word of caution is given about the use of ‘unsupervised’ which is sometimes extended from its proper meaning as (roughly): learning from data but without the provision of correct target structures for that data’. In that sense, machine translation (MT) would be learned in a supervised manner when exposed only to parallel (translated) corpora, but if MT could be learned (which is highly doubtful and never yet demonstrated) only from a range of monolingual texts in various languages, that would be unsupervised learning. This issue is sometimes muddled up with that of provided corpora marked up in a sophisticated manner for supervised learning: such as marking each content word with its appropriate word-sense against some dictionary sense list, so as to train word-sense disambiguation (WSD). This is a highly labour intensive task and researchers yearn for easier methods of acquiring training data, though this is not the same as lack of supervision. For example, those same parallel language texts could be used to tag the senses of one word against those in another but this would not be unsupervised learning, just cheap supervision (e.g. if ‘duty’ were tagged by occurring opposite translations Œdevoir vs. ‘impot’ in French). To put the matter as Manning and Schuetze do (1999): unsupervised learning is a clustering task and supervised learning a classification one.
There is still much mileage, to judge from bulletin board postings, of new forms of the old opposition of generative and corpus linguistics, but I would suggest that, if we leave aside the wholly unreformed Chomskyans, where any kind of objective evidence is concerned, then that opposition may be correlated with access to the two sorts of corpora I distinguished above: corpora and metacorpora (i.e. electronic dictionaries and thesauri). I then describe a small piece of recent work with a colleague (1998) where we first calculated what kinds of links of WSD to part-of-speech tagging were possible from the distribution of homographs in LDOCE. We then checked against a small text corpus which we had hand tagged with senses to see what level of facts we actually found. The distinction between the two stages, computation over a lexicon then a corpus, corresponds roughly to a Gedankenexperiment and then an experiment, and also to a form of generative vs. corpus linguistics distinction where generativity is expressed in a dictionary or metacorpus.
One can see a stronger form of this by moving from the WSD to the issue of novel senses in text and asking the question as to whether novel sense detection can ever be linked to corpus methods: can it even in principle be marked up for in training and test texts by humans? In some sense the answer must be no by definition, although we might seek the presence of a new sense where a tagger had been unable to assign any existing sense to a word. It might be interesting to consider this within the Generative Lexicon paradigm (GL, Pustejovsky 1995), which is very much a generative approach in the sense of being (a) prescient about real corpora and (b) attempting to compact lexicons by full intuitive compactions of their content. We ask how much of novel sense, with which Pustejovsky claims to deal, is of this type and cite a recent computation over LDOCE by Kilgarriff (2001) that tries to show that virtually no novel sense is captured by GL. Is this a real or a Gedanken experiment? Is it fair to GL?
Finally, we look at a different, brute force approach, to attempting to locate novel sense in corpora -- an area hitherto unexplored we believe -- this time unsupervised, by asking how many of the agent noun + verb combinations in the BNC are unique -- uniqueness, or at least very low frequency being a place where novel sense might be expected to emerge. We then look at these numbers -- which are surprisingly large, at least to me -- and ask which of them are comprised of nouns and verbs which are themselves frequent in the BNC. I then display what we get and ask if it has any significance for the issue of constraints on possible levels of WSD.
The determiner slot in English interrogative Noun Phrases allows for three possible occupants, namely which, what and whose, and their variants ending in -ever: whichever, whatever and whosever. In this paper we will be looking only at the first two of these, which and what, in structures like the following: Which films do you like? / What films do you like? These expressions are close in meaning, so the question arises as to what factors influence the choice of one over the other determiner. Our aim in this paper is twofold: first we will demonstrate how interrogative NPs such as those exemplified above can be retrieved from the British component of the International Corpus of English (ICE-GB), using Fuzzy Tree Fragments (FTFs). The second aim of the paper is to show how the search results can be used to investigate what determines the choice of determiner in NPs like those shown above.
Our findings show that although the accounts found in the grammar books regarding the choice of which or what as interrogative determiners seem to be valid in most cases, some modifications are necessary. We have found that while the use of what implies a choice of answers from a seemingly unlimited set, the set may in fact have an upper or lower bound, or both. We have also found that pragmatic factors can play a role, such as speaker and hearer expectations. Finally, there are a number of instances that simply defy the rules. Given the examples from ICE-GB that we examined, the standard accounts of the use of which and what as interrogative determiners are probably best regarded as expressing a tendency, which may be influenced, or superseded, by a variety of factors.
Hajar Abdul Rahim & Harshita Aini Haroon (Universiti Sains Malaysia, Malaysia)
Odlin (1989) differentiates two kinds of language mixing: borrowings from a second language into the native language and code-switching, a systematic interchange of words, phrases, and sentences of two or more languages. Whilst Odlin considers borrowings as a kind of language mixing, Hatch & Brown (1995) stress the need to distinguish between borrowings and language mixing. In borrowing, the words become part of the language used by speakers of that language, as though they were native words such as garage (French) and pizza (Italian). In mixing and switching, the words are momentarily borrowed by individual speakers in order to create effects.
Code-switching/language mixing between Malay and English is prevalent in Malaysian English, a variety of English spoken by Malaysians. Another important feature in Malaysian English is borrowing. This paper, based on a corpus study of Malaysian English, attempts to describe the language mixing patterns and the linguistic (semantic and pragmatic) features of borrowings (from Malay) in Malaysian English.
The study analysed a set of 11 Malay words extracted from a 60,000-word Malaysian English corpus. The corpus comprises 6 text-types (2 spoken and 4 written). The spoken texts are classroom lessons and broadcast news whilst the written texts are student essays, press news reports, press editorials and non-academic writing. Each text type was compiled in 5 files of 2,000 words each. The WordSmith Tools (1998) package was employed to locate the Malay words in the corpus. A wordlist was initially generated from which the 11 Malay words were extracted.
The words analysed are balik kampung, changkuls, ikan kembong, jahanamkan, malu, nabi, padi, pahala, rakyat, samsu and ummah. The words and their meanings in English are listed below:
The root form jahanam which means ‘ruin/ complete damage or destruction’ is borrowed from Arabic. The word jahanamkan means ‘to ruin or damage completely’.
Rice which is growing or still in the husk / paddy.
‘Reward’ in the religious context.
‘The people’ or ‘citizens of a state’.
The 11 items analysed above turned out to be words that occur in the written texts only. An analysis of the meanings of the words suggests that the Malay items could be categorised as follows:
1. 1. Words that have no English equivalent.
2. 2. Words that have a close English equivalent.
3. 3. Words that have an English equivalent (same semantic content).
The words in the first category are balik kampung, samsu and ummah. They contain cultural information that is peculiar to the Malaysian social/cultural context: balik kampung- peculiar initially to Malays, it has become a Malaysian phenomenon particularly around holidays. samsu- this locally produced alcohol is popular among certain socio-economic groups; ummah– this is used particulary to refer to people of the Islamic faith.
The words that fall into the second category are ikan kembong, jahanamkan, malu, and pahala. Although there are near equivalents that sufficiently provide the basic semantic content of the words, the inclination to use the Malay words is due to various reasons. The use of ikan kembong instead of ‘mackeral’ is a case of familiarity of term within the social group. The use of pahala instead of ‘reward’ is very likely motivated by the semantic information [+religious] and [- material] inherent in pahala, but which is not included in the meaning of ‘reward’.
Words that fall into the third category, i.e. those that have English equivalents, are changkul, nabi, padi, and rakyat. The insistence on using Malay words when there are English equivalents may be attributed to the language choice of users and the interpretation of the message. Although the English equivalents have the same conceptual meaning, the use of the Malay words is possibly motivated by the pragmatic effects the connotations of the words have in the interpretation of the message.
The study has used a limited number of words extracted from a small corpus. One needs to consider this limitation before any conclusions could be made on the 1) language mixing patterns and 2) the linguistic features of borrowings in Malaysian English. However, based on the data obtained and the assumption that borrowing is a long-term process within the language of a social group and mixing is a momentary individual phenomenon, it is suggested that words that have no English equivalent are borrowed in Malaysian English. The words that have the English equivalent and near equivalent are used as a code-switching/mixing strategy.
In all three categories of words, the semantic content of the Malay words is not altered in the context of Malaysian English. The semantic content is maintained as it contributes towards the interpretation of the message. The pragmatic information of the words in all categories is also maintained in the Malaysian English context. However, the pragmatic content seem to be more necessary in the cases of words that have the English equivalent.
Hatch, E. and Brown, C. (1995) Vocabulary, Semantics, and Language Education. Cambridge University Press, Cambridge.
Odlin, T. (1989) Language Transfer. Cambridge University Press, Cambridge.
WordSmith Tools Version 3.0. Oxford University Press.
Eric Atwell (University of Leeds, United Kingdom)
The goal of project ISLE (Interactive Spoken Language Education) was to exploit available speech recognition technology to improve the performance of computer-based English language learning systems. The ISLE project also collected a corpus of audio recordings of German and Italian learners of English reading aloud selected samples of English text and dialogue, to train the speech recognition and pronunciation error-detection modules. Speech recordings were collected from non-native, adult, intermediate learners of English: 23 German and 23 Italian learners. In addition, data from two native English speakers was collected for test calibration purposes. The corpus contains 11484 utterances; 1.92 gigabytes of WAV files; 17 hours, 54 minutes, and 44 seconds of speech data. The corpus is based on 250 utterances selected from typical second language learning exercises. It has been annotated at the word and the phone level, to highlight pronunciation errors such as phone realisation problems and misplaced word stress assignments.
In addition to the blocks of individual speaker data, we created five pseudo-speaker blocks of data by selecting some utterances covering all speakers, in order to be able to check inter and intra-annotator consistency. Overall, agreement rates were low: at best, annotators agreed in only 55% of cases when deciding where and what an error is. Even localisation of the error alone, deciding where the error is but not what the correction should be, shows at best a 70% agreement between annotators. In some cases this was because annotators flagged errors in the same word but not the same exact location (phoneme). Given the poor inter-annotator agreement on the exact location and nature of errors, the target one might reasonably set for diagnosis programs should be limited to only those errors which annotators agree on; this applies not only to the ISLE system but to other pronunciation correction systems.
Statistics extracted from the error-annotated corpus allow us to see which are the most common sources of English pronunciation errors for native speakers of Italian and German. For both Italian and German native speakers, we have empirical evidence on which are the most difficult phones and which phones account for most errors (equivalent to the type/token distinction in corpus frequency counts), and which words account for the most errors. The Italian speakers made an average of 0.54 phone errors per word with a standard deviation of 0.75, while the Germans made an average of 0.16 phone errors per word with a standard deviation of 0.42. This difference may be partly due to the greater phonological similarities between German and English than between Italian and English. Examples of pronunciation errors at each level are given, with an indication of whether these are expected (owing to L1 interference and attested in the EFL literature) or unpredictable/idiosyncratic.
We welcome corpus re-use by other researchers, who can acquire a copy (on 4 CDs) from ELDA. At the end of the project, system development stopped at the Demonstrator stage, and future prospects for migration to a commercial ELT package are uncertain; however, we hope that the ISLE Corpus may be a useful achievement of the project.
This paper reports on a collaborative research project; I gratefully acknowledge contributions of number of collaborators, principally: Wolfgang Menzel, Dan Herron, and Patrizia Bonaventura, University of Hamburg (Germany); Steve Young and Rachel Morton, Entropic Cambridge Research Laboratory Ltd. (Cambridge, UK); Jurgen Schmidt, Ernst Klett Verlag (Stuttgart, Germany); Paulo Baldo, Dida*el S.r.l. (Milan, Italy); Roberto Bisiani and Dan Pezzotta, University of Milan Bicocca (Italy); and last but definitiely not least, my colleagues at the University of Leeds (UK), Peter Howarth and Clive Souter. This research was supported by the European Commission under the 4th framework of the Telematics Application Programme (Language Engineering Project LE4-8353). The corpus is distributed for non-commercial purposes through the European Language Resources Distribution Agency (ELDA).
I am particularly endebted to Wolfgang Menzel for setting up and leading the ISLE project; and to Uwe Jost for proposing Leeds University as a contributor to the project. I am also grateful that the ISLE project as allowed me to achieve a long-standing ambition to contribute my own linguistic data to a Corpus!
Atwell, E. (1999) The Language Machine. British Council, London.
Atwell, E., Howarth, P., Souter, C., Baldo, P., Bisiani, R., Pezzotta, D., Bonaventura, P., Menzel, W., Herron, D., Morton, R., and Schmidt, J. (2000) User-Guided System Development in Interactive Spoken Language Education, Natural Language Engineering Journal 6(3-4):229-241, Special Issue on Best Practice in Spoken Language Dialogue Systems Engineering, Cambridge University Press, Cambridge.
Eisen, B., Tillmann, H., and Draxler, C. (1992) Consistency of Judgements in Manual Labelling of Phonetic Segments: The Distinction between Clear and Unclear Cases. In: Proceedings of ICSLP’92: International Conference on Spoken Language Processing, pp. 871-874.
Herron, D., Menzel, W., Atwell, E., Bisiani, R., Daneluzzi, F., Morton, R. and Schmidt, J. (1999) Automatic localization and diagnosis of pronunciation errors for second language learners of English. In Proceedings of EUROSPEECH'99: 6th European Conference on Speech Communication and Technology, vol. 2, pp.855-858. Budapest, Hungary.
Hunt, J. (1996) The Ascent of Everest. Ernst Klett Verlag, English Readers Series, Stuttgart.
Menzel, W., Atwell, E., Bonaventura, P., Herron, D., Howarth, P., Morton, R., and Souter, C. (2000) The ISLE Corpus of non-native spoken English. In Gavrilidou, M., Carayannis, G., Markantionatou, S., Piperidis, S. and Stainhaouer, G. (eds) Proceedings of LREC2000: Second International Conference on Language Resources and Evaluation, vol. 2, pp.957-964. Athens, Greece. Published and distributed by ELRA - European Language Resources Association.
Young, S., Kershaw, D., Odell J, Ollason, D., Valtchev, V. and Woodland, P. (1999) The HTK Book 2.2. Entropic, Cambridge.
Power, K., Morton, R., Matheson, C., and Ollason, D. (1996) The Graphvite Book 1.1. Entropic, Cambridge.
Thomas, J. (2001) Negotiating meaning: a pragmatic analysis of indirectness in political interviews. Invited plenary paper, Corpus Linguistics 2001 Conference, Lancaster University, UK.
Weisser, M. (2001) A corpus-based methodology for comparing and evaluating different accents. In: Rayson, P., Wilson, A., McEnery, T., Hardie, A. and Khoja, S. (eds) Proceedings of the Corpus Linguistics 2001 Conference, pp.607-613. UCREL: University Centre for computer corpus Research on Language, Lancaster University, UK.
Roumiana Blagoeva (Sofia University, Bulgaria)
As part of a wider investigation of cohesion in learner writing this paper discusses some aspects of personal reference in argumentative essays of non-native advanced learners of English. The study is comparative, based on data drawn from three electronic corpora of about 100,000 words each: a learner corpus of argumentative essays written by Bulgarian university students of English language and literature, compiled within the framework of the ICLE project for collecting interlanguage data from advanced learners of English of different mother tongue backgrounds; a native-speaker written corpus of authentic English non-fiction texts, which are used as teaching materials with the contributors to the learner corpus; and a relevant corpus of Bulgarian texts from a variety of sources.
The work reported on here is concerned with the occurrences of personal pronouns in the learner written production and their function as reference items for the construction of coherent texts. On the basis of comparisons with the target language and the native language of the learners an attempt is made to offer explanations of the learners’ preferences for certain discourse patterns.
The frequency lists produced for all personal pronouns in the learner and native corpora showed a striking overuse of all items in the learner production in English. The greatest differences, however, were observed in the use of third person singular personal pronouns in the nominative (subjective) case. The analysis focuses on the different functions and uses of it where the ratio between its occurrences in the native and the learner corpora is 1:2,5.
To explore the reasons underlying this phenomenon similar searches were applied to the Bulgarian corpus with the aim of revealing characteristic features of Bulgarian personal pronouns and their possible influence on the acquisition of English by Bulgarians. Here it seems pertinent to mention some similarities and differences between the role of personals as cohesive devices in the TL and NL of the learners. As far as textual relations are concerned, pro-forms in both English and Bulgarian behave in a similar way. Being items void of conceptual meaning they function in the surface text as elements indicating that information about their meaning is to be retrieved from elsewhere: either from the situation of the communication act thus relating exophorically to entities in the world outside the text, or from the text itself when they refer endophorically to preceding or following items expressing anaphoric or cataphoric reference respectively. Two major dissimilarities, however, exist between the systems of personals in English and Bulgarian, which are most prominent in the third person singular. They arise from the different expression of the category of gender and the opposition animate/inanimate in the two languages, and from the inflectional character of Bulgarian which allows the omission of pronouns. While in English a non-personal inanimate item denoting an object or a thing is referred to most often with the pronoun it, the choice of a pronoun in Bulgarian for the same inanimate entity will depend on the gender of the noun it co-refers with. Therefore, the equivalents of the English it could be той [toj] (masculine, animate/inanimate), тя [tja] (feminine, animate/inanimate), and то [to] (neuter, animate/inanimate).
Furthermore, all Bulgarian nouns and verbs are marked for gender and number through special suffixes and it is quite natural for speakers of Bulgarian to avoid repetitions of pronouns whenever possible. The explicit mention of personal pronouns in the nominative is obligatory only if we need to disambiguate certain contexts or to express emphasis.
In all other cases it is a matter of personal choice on the part of the speaker to use or not to use pronouns in the surface text. Such omissions are treated by most authors not as ellipsis but as implicit/explicit expression of the pronoun.
Each of the types of referential ties mentioned above is examined separately in terms of quantity and quality and sample sentences from the three corpora are discussed. The differences between Bulgarian and English personals are taken into consideration and it is compared with its three equivalents only when they express relations similar to those of it. The processing of the data at this stage is done semi-automatically. The cases of implicit expression of third person pronouns are not included in the investigation, first, because samples cannot be retrieved automatically from the corpus, and more importantly, because they cannot indicate L-1 induced overuse of these items. On the contrary, the lack of formally expressed pronouns in Bulgarian would suggest underuse of these items by Bulgarian learners of English.
It is quite obvious, then, that the overuse of personals by learners of English is hardly due to NL interference. The idea that there exist culture-specific patterns of writing would also be of little value in providing a plausible explanation of the differences between learner and native writing. At this stage of the investigation into cohesion in learner writing one of the possible causes could be sought in a communication strategy common to many advanced second language learners, namely that at a certain point of FLA they feel confident enough to communicate in the foreign language and “stop learning” in the sense that they tend to stick to some language patterns fossilized at an earlier stage of learning. Further corpus-based research in this area is likely to enhance our understanding of intuitive judgements about learner production and point to effective ways of developing interlanguages.
Mia Bostrom Aronsson (Göteborg University, Sweden)
Learner writing is known to differ from native speaker writing in several ways, for instance in terms of frequency of certain words or structures. One type of construction that is frequently overrepresented in Swedish advanced learners’ written English is different types of cleft constructions. These are a type of focusing device used to manipulate the thematic structure and the information structure of a text. The overrepresentation of these constructions in Swedish advanced learner writing is probably caused by several different factors. This paper discusses some ways in which Swedish advanced learners’ use of it-clefts and pseudo-clefts, as in (1) and (2), differs from native speaker use, some possible underlying causes of the differences, and what effect the learners’ use of these constructions may have on their texts.
(1) It was Tom who offered Sue a sherry (Collins 1991:3)
(2) What Tom did was offer Sue a sherry (Collins ibid)
It-clefts and pseudo-clefts are flexible constructions that can be used to rearrange the order of the sentence elements to make an element focal and/or thematic, which would not be so in a regular declarative sentence. This is particularly useful in writing, where prosody is not marked (Quirk et al. 1985: 1384). As can be seen in example (1), for instance, the cleft construction places additional focus on the subject Tom, which would not be in focus in a regular declarative sentence (Tom offered Sue a sherry Collins 1991: 4). Example (3), which is an authentic example from a native speaker text, illustrates how the use of an it-cleft can contribute to the organization of the text in a paragraph in that the it-cleft makes it possible to focus on the subject these people…, which forms a cohesive link to the previous sentence by means of the anaphoric reference to fight promoters and fighter managers, at the same time as the focused noun phrase is placed early in the sentence.
(3) Naturally, there is the argument for keeping boxing. As I said before, it has
developed into an extremely lucrative sport with millions of pounds being offered for the elite to fight. The fight promoters and fighter managers will all be receiving large sums of money. It is these people who are directly involved with the sport who would defend it greatly. (LOCNESS-A-level-Boxing-B10, italics added)
In addition to the text organizing function of clefts, these constructions are also associated with an exclusiveness implicature and an existential presupposition, which entail that in an example such as (3), the cleft construction expresses exclusively who would defend the sport and that it is a fact that someone would defend it (see further Collins 1991: 69ff; Huddleston 1984: 465ff; Johansson 1996: 129ff).
This study compares the use of cleft constructions in argumentative essays written by native Swedish students in their second year of university studies of English with the use of these constructions in argumentative essays produced by native speakers of English. The learner writing consists of material from the Swedish component of the International Corpus of Learner English (ICLE), whereas the native speaker material is taken from the LOCNESS corpus (Louvain Corpus of Native English Essays). The study looks into how the learners’ use of cleft constructions differs from native speakers’ use as regards the form of the constructions and the function of the examples in their context. The analyses of the form of the pseudo-clefts and of which elements are placed in focus of pseudo-clefts indicate, among other things, that the differences between the learner examples and the native speaker examples may reflect differences as regards the argumentative styles of learners and native speakers and that these differences may contribute to the high frequency of cleft constructions in Swedish advanced learner writing.
A study of the learner and native speaker examples in their context indicates that the learner examples often appear unmotivated in their context, whereas this is not common in native speaker writing. For instance, the learner examples often emphasize elements that do not need to be emphasized, judging from the context. The unmotivated use of cleft constructions may give an implication of exclusiveness which is not relevant in the context. This may have a negative effect on the coherence of the text. Moreover, the learner examples also place elements as marked themes even though there is no need for the particular element to be thematic. Rather, the fact that it is made thematic may have a negative effect on the thematic development of the text. Thus the study of cleft constructions in their context in Swedish advanced learner writing and native speaker writing indicates that the learners’ use of cleft constructions reflects the fact that Swedish advanced learners have problems with the distribution of information in their texts.
Collins, P. C. (1991) Cleft and Pseudo-Cleft Constructions in English. Routledge, London.
Huddleston, R. (1984) Introduction to the Grammar of English. Cambridge University Press, Cambridge.
Johansson, M. (1996) Contrastive data as a resource in the study of English clefts. In Aijmer, K., Altenberg, B., and Johansson, M. (eds). 1996. Languages in Contrast: Papers from a Symposium on Text-based Cross-linguistic Studies. Lund 4-5 March 1994. Lund University Press, Lund, pp. 127-150.
Quirk, R., Sidney G., Geoffrey L., and Svartvik, J. (1985) A Comprehensive Grammar of the English Language. Longman, Harlow.
Nicholas Brownlees (University of Florence, Italy)
The ZENcomp corpus is an addition to the full ZEN (Zurich English Newspapers) corpus. Whereas the latter covers the period 1661-1800, ZENcomp incorporates the first four decades of English newspaper publication. These years (1620-1660) saw English newsbooks, as periodical news publications were then called, develop from rough, badly translated newssheets to quite subtle, well-informed news texts. The first decade to be included in the ZENcomp corpus is that of the Civil War years (1641-1650). Like its parent corpus, ZENComp includes about 140,000 words of newspaper text for each of the four decades (1620-1660) under review. Such a figure is relatively ample for much of the seventeenth century, where newspaper publication is generally quite uniform and frequently based around one central publication (such as the London Gazzette for the period 1665-1688) but decidedly more restrictive when measured against the plethora of highly diverse publications of the Civil War period.
The years in question were as unique in seventeenth century newspaper history as they were in English politics and society. Periodical news publications had begun in England in 1620, but it was only in 1641, with the breakdown of royal authority, that they began to report domestic matters. Furthermore, as centralised political power waned, and political and religious controversy deepened, the number of publishers tempted to enter the world of newsbook publication increased. In no other decade of the seventeenth century is the English press so free, politically relevant, numerically significant and stylistically heterogenous. As can be imagined, for the corpus creator such a situation is both highly stimulating and very problematic. Out of the hundreds of separate publications available which ones should be selected for the corpus and why1?
Hard choices need to be made and selection necessarily depends on thematic and linguistic objectives. The first question is whether to include only a part or all of separate newspaper publications. As Civil War newsbooks were generally between 8 to 12 quarto pages long, it was decided that the advantages of having complete newspapers in the corpus overrode a decrease in thematic and stylistic representation. The 1641-1650 period will only include about 45 different newspaper publications but at least these will be complete texts with all the consequential advantages for macro- as well as microlinguistic analyses.
The next important question concerns the kind of newspapers selected for inclusion. Civil War newspapers range from one political extreme to another, one register to another, and while some titles survived a number of years others died a few days after birth. How inclusive should a Civil War corpus attempt to be? Furthermore, apart from the nature of the publication, how much importance, if any, should be given to the date of publication? The decade in question was momentous, but some dates particularly stand out - for example, the Battle of Naseby in 1645 and the execution of Charles I in 1649. Decisions have to be made as to whether such dates should be highlighted, avoided, or regarded as inconsequential. In the progress report I shall expand on these sampling issues and indicate some of the principal linguistic characteristics a corpus of ZENcomp's dimensions can highlight.
The English language features some verbal constructions, which can variously be treated as multi-word lexical units or as syntactic sequences with a more or less loose connection between the individual elements. Among them are verb-preposition sequences, some of which have been classified as prepositional verbs, i.e. lexemes in their own right (e.g. Quirk et al. 1985, Vestergaard 1977, Diensberg 1990). However, neither the precise extent of membership in this class nor even its very existence (e.g. Huddleston 1984, Götz/Herbst 1989) has remained undisputed.
This paper looks at verb-preposition sequences in the BNC Sampler in order to establish both the outer limits of a class 'prepositional verbs' as well as its possible internal sub-divisions. As verb-preposition sequences are extremely frequent, the following somewhat restricted search procedures have been used: (i) base form of verb (VV0), third person verb (VVZ) and past tense form of verb (VVD) followed by general preposition (II), either immediately, and with one/two words intervening, (ii) past participle (VVN) followed immediately by general preposition as well as (iii) general preposition followed by whom, who, whose, which and what. This approach is intended to yield, apart from straightforward verb-preposition sequences, instances of prepositional passives/preposition stranding, pied-piping and inserted adverbials, all of which features have played a role in the discussion of this construction so far.
The search results were then weeded out according to the following principles. If the preposition is to be more closely connected to the verb, which is a precondition for prepositional verb status, the noun following it must be an independent role-playing participant in the clause. Thus all cases where the preposition introduces a phrase functioning as a circumstantial element of extent and location in time and space, manner (means, quality and comparison), cause, contingency, accompaniment, role and angle (cf. Halliday 1994: 151) were excluded. Furthermore, instances of phrasal-prepositional verbs (e.g. look forward to), verb-noun combinations (e.g. give way to, fall in love), and cases where a direct object was present or possible (as intervening word, e.g. turn N into N) were also discarded. Cases of preposition stranding, on the other hand, were always retained regardless of their behaviour regarding the criteria just mentioned.
The remaining examples are then examined with regard to the following aspects:
- the exact nature of the noun following the verb-preposition sequence. The fact that the noun is a role-playing participant need not necessarily imply that the verb and preposition indeed form a close unit. Verbs which can optionally omit their direct object (e.g. write to N) are of interest in this respect, for example.
- collocational fixity, i.e. the (lack of) commutability of the verb and especially the preposition making up the sequence. Here, verbs with variable prepositional usage (e.g. consist of/in), and verbs occurring with or without the preposition (e.g. decide/decide on) will have to be looked at more closely. The stronger the collocational bond, the more likely the existence of a prepositional verb.
- syntactic tests which can be helpful in establishing either an SVA- or an SVO-analysis for sentences with verb-preposition sequences, such as look into a problem, live on little money, and sleep in a bed. Among them are the possibility of the passive-transformation, the extent of preposition stranding or of pied-piping, and the insertion of adverbials between verb and preposition. None of them have been regarded as conclusive so far. However, the occurrence of these syntactic features may show up some valid tendencies.
- semantic opacity/non-compositionality, in particular whether this should be taken as a criterion at all. Here, the semantic content of the preposition and its concrete vs. abstract use (cf. arrive at the station/arrive at a conclusion) will play a role. Many potential prepositional verbs seem to be (fairly) 'literal' (e.g. insist on, believe in), while the more opaque ones (e.g. set about, come by) have even been classified as phrasal verbs by some researchers (e.g. Dixon 1992) – an approach that ignores the syntactic differences between these two verb types.
- the type of verbs occurring in prepositional verbs. It might be relevant whether the verb is of Romance or Germanic origin and how integrated it is into the core of the lexicon. Combinations such as look into, come by, call on have a different flavour from, e.g. rely on, relate to and insist on, with the former being more versatile and semantically more akin to phrasal verbs.
- the use of preposition as a pure affix/casemarker, as in look at a picture, or as a meaning-modifying element, e.g. play at politics, know about something. While the preposition as part of a prepositional verb has been seen to have very little or no meaning of its own independent of the verb and stand in no opposition to other prepositions, in the latter use the preposition does make a semantic contribution. It modifies the nature of the action denoted by the verb, not the meaning or status of the following noun.
This paper will argue that, while the criteria can be conflicting in cases, it is nevertheless possible to distinguish a class of prepositional verbs, with the status of the following noun, the nature/indispensability of the preposition and the collocational bond probably being the primary criteria. Furthermore, three types of prepositional verbs are proposed, namely (i) verbs which need the prepositional affix to introduce their object and where the preposition is semantically unimportant (e.g. rely on), (ii) verbs which enter into a close semantic relationship with the preposition, producing an idiomatic and/or opaque combination (e.g. look into) and (iii) verbs which in some uses add a preposition which produces a semantic modification (e.g. play at) (cf. also Goyvaerts 1973, for whom only (ii) are clearly prepositional verbs).
Diensberg, B. (1990) A syntactic analysis of English prepositional verbs, Tromsø Studies in Linguistics 11: Tromsø Linguistics in the Eighties: 85-109.
Dixon, R. M. W. (1992) A New Approach to English Grammar, on Semantic Principles. Clarendon Press, Oxford.
Götz, D. and Thomas, H. (1989) Language description and language teaching: The London School and its latest grammar, Die Neueren Sprachen 88: 220-235.
Goyvaerts, D. L. (1973) Some observations about the verb+particle construction in English, Revue de Langues Vivantes: 549-562.
Halliday, M.A.K. (1994) An Introduction to Functional Grammar. 2nd ed. Arnold, London.
Huddleston, R. (1984) Introduction to the Grammar of English. Cambridge University Press, Cambridge.
Quirk, R., Greenbaum, S., Leech, G. and Svartvik, J. (1985) A Comprehensive Grammar of the English Language. Longman, London.
Vestergaard, T. (1977) Prepositional Phrases and Prepositional Verbs. A Study in Grammatical Function. Mouton, The Hague/Paris.
In his well-known 1991 article, Kjellmer puts forward the hypothesis that learners of English, unlike native speakers, tend to construct utterances from individual words, rather than sequences of words, that their ‘building material is individual bricks rather than prefabricated sections’. However, findings from a corpus-driven study of ‘repetitive phrasal chunkiness’ (Lancashire 1996) in native and learner speech and writing (De Cock 2000) seems to contradict this hypothesis, revealing that the learners used more continuous recurrent n-grams (i.e. sequences of n orthographic words) than comparable native speakers, both in speech and more especially in writing.
The aim of the present study is to provide a qualitative follow-up to this quantitative study of continuous recurrent n-grams through an analysis of fully comparable corpora of native speaker and learner informal interviews and native speaker and learner non-professional argumentative essay writing. I set out to investigate the nature of recurrence in these specific genres and to determine the extent to which native speakers and learners use similar recurrent sequences.
As is clear from the lists of automatically extracted recurrent n-grams, not every frequently recurring sequence can be considered as a well-established phraseological expression. Sequences such as there’s a, in the, the the the, but er and I don’t know if are all high frequency sequences without being ‘phraseological expressions’ as such. As Moon (1998) points out, recurrence does not automatically indicate phraseological status. In other words, there are different types of recurring sequences.
A model for the description of the various kinds of continuous n-grams in the corpora is put forward and the proportion of these different kinds of sequences is assessed for each combination length (both in terms of types and tokens) to find out whether and to what extent the same types of recurrent n-grams are used in the spoken and written genres under investigation and in the native and learner varieties. Particular attention is paid to the question of whether learners’ overuse of recurrent n-grams in speech and especially in writing noted in De Cock (2000) can be attributed to an overuse of one or more specific (sub)categories.
Among the major types of recurrent sequences that can be distinguished, there are, at one level, phraseological multi-word lexical units, which can be seen to display varying degrees of non-compositionality, restricted collocability and restricted flexibility. They are often presented as typically native-like (Pawley and Syder 1985, Granger 1998) and as particularly common in speech (Aijmer 1996, Altenberg and Olofsson 1990). They include, for example, the complex preposition because of, the complex connectors of course and on the other hand or the comment clauses I mean and you know. I will attempt to discover whether these recurrent phraseological expressions make up a larger proportion of the recurrent strings in the native than in the learner corpora and if so, if this tendency is more significant in the spoken than in the written corpus.
Beside these multi-word lexical units, there is a series of structurally complete multi-word sequences (phrases and clauses), the majority of which are not strictly phraseological but which, just like phraseological multi-word lexical units, can nevertheless be labelled as ‘preferred building blocks’ (Altenberg 1998). They can be seen to reflect what Béjoint (2000) calls ‘tendencies in the encoding of text’, tendencies which he regards as part of the ‘mastery of language’. Strings like I don’t think so, most of the time or away from home are arguably also part of the idiomaticity of English taken in the wide sense, i.e. in the sense of Pawley and Syder’s (1983) ‘native-like selection’ or of Sinclair’s (1991) ‘idiom principle’.
At another level, there are structurally incomplete sequences or ‘phrase or clause fragments’ such as of the, because of the, one of the, but I, there is a, the high frequency of most of which is largely due to the very high frequency of the words which compose them. Some of these sequences can be described using terminology/categories from Altenberg’s (1998) investigation of recurrent word combinations in the London-Lund Corpus of Spoken English (cf. ‘multiple clause constituents’ and more especially ‘frames’, ‘onsets’ and ‘stems’).
At yet another level, there are what we could call ‘speech-specific n-grams’, i.e. n-grams that contain what have been referred to as performance errors. ‘Speech-specific n-grams’ contain reduplication (the the, I I I) and/or hesitation items such as er or erm (and er, er the the). Although it may be tempting to dismiss phrase or clause fragments and speech specific n-grams as not worthy of study, a preliminary analysis reveals that it is important to include them in our description as they are part and parcel of the phenomenon of recurrence. The study reveals, among others, that learners’ overuse of recurrent n-grams in speech (De Cock 2000) is in fact largely due to an overuse of speech specific n-grams.
The question arises as to whether the apparent contradiction between Kjellmer’s statement and the overuse of recurrent sequences noted in De Cock (2000) still holds when speech-specific n-grams and some types of phrase/clause fragments are discarded.
Altenberg, B. and Eeg-Olofsson, M. (1990) Phraseology in Spoken English: Presentation of a Project. In Aarts, J. and Meijs, W. (eds) Theory and Practice in Corpus Linguistics. Rodopi, Amsterdam/Atlanta, pp. 1-26.
Béjoint, H. (2000) Modern Lexicography: An Introduction. Oxford University Press, Oxford.
De Cock, S. (2000) Repetitive phrasal chunkiness and advanced EFL speech and writing. In Mair, C. and Hundt, M. (eds) Corpus Linguistics and Linguistic Theory. Papers from the Twentieth International Conference on English Language Research on Computerized Corpora (ICAME 20) Freiburg im Breisgau 1999. Rodopi, Amsterdam and Atlanta, pp. 53-68.
Granger, S. (1998) Prefabricated patterns in advanced EFL writing: Collocations and formulae. In Cowie, A. P. (ed.) Phraseology: theory, analysis and applications. Oxford University Press, Oxford.
Kjellmer, G. (1991) A mint of phrases. In Aijmer, K. and Altenberg, B. (eds) English Corpus Linguistics. Longman, London/New York, pp. 111-127.
Lancashire, I. (1996) Phrasal Repetends in Literary Stylistics: Shakespeare’s Hamlet III.1. In: Hockey, S. and Ide, N. (eds) Research in Humanities Computing 4. Selected Papers from the ALLC/ACH Conference, Christ Church, Oxford, April 1992. Calderon Press, Oxford, pp. 34-68.
Roberta Facchinetti (University of Verona, Italy)
0. 0. Introduction
Over the past few decades, scholars have written extensively on English modality and on modal verbs in particular. A quick overview of only a few studies carried out and/or published in the year 2000 shows how research has thrived particularly in the field of Present-day English modal verbs (Facchinetti 2000, Leech forthcoming, Palmer forthcoming, Papafragou 2000a, 2000b, Winford 2000) and of their historical development (Krug 2000, Vihla 2000, Myhill forthcoming) MAY is among the most targeted modals, due to the dramatic semantic changes it has undergone through the centuries, but also to the values and functions it conveys in Present-day English.
Faithful to this long-lasting research tradition, and in the hope of contributing to enlightening some still shaded tessellas in the mosaic of MAY, I will also carry out a corpus-based study of this modal, with the aim of charting its distribution and semantic/pragmatic values in British English.
To do so, I will analyse the British Component of the International Corpus of English (ICE-GB), which contains a total of 1 million running words distributed among a wide range of textual types (Greenbaum 1996).
1. 1. Quantitative distribution
A total of 1219 instances of MAY have been recorded in the corpus under scrutiny. Their distribution widely confirms the generally acknowledged view the MAY is employed with much higher frequency in the written medium than in the spoken medium (frequency per ten thousand words: 6,5 in speech vs. 19 in writing). The formality constraints and the semantic/pragmatic values of the modal strongly affect its distribution among the textual categories represented in the corpus, the most blatant discrepancies being the following:
· · within public dialogue: 'broadcast interviews' (4,1) vs. 'parliamentary debates' (12,3) and 'legal-cross examinations' (13,2);
· · within monologue: 'spontaneous commentaries' (3,3) vs. 'legal presentations' (13,8);
· · within non-professional writing: 'untimed student essays' (15) vs. 'student examination scripts' (22,6);
· · within correspondence: 'social letters' (8,7) vs. 'business letters' (16,1);
· · within printed texts: 'instructional writing' (38,5) vs. 'news reports (12,8).
Other sociolinguistic variables, pertaining to the age, the gender, and the level of education of the speaker/writer, have been taken into consideration in the present analysis, but only the level of education has yielded statistically relevant results with reference to the distribution of MAY, since speakers/writers with a university level of education appear to use MAY more frequently than people who have a secondary school level of education.
2. 2. Semantic and pragmatic values
Unsurprisingly, MAY is mostly associated with epistemic modality, but deontic and dynamic values have also been recorded, though to a much more limited extent, as shown in Figure 1:
Figure 1: semantic values of MAY in ICE-GB
2.1. Epistemic modality: possibility
A total of 746 occurrences (61%) express 'epistemic possibility' and are particularly frequent in the following textual types:
· · private conversations
· · business transactions
· · spontaneous commentaries
· · broadcast news/talks/discussions
· · social letters
· · academic writings - humanities
· · non-academic writings - humanities, social sciences
· · press news reports/editorials
The large majority of these instances are positive, as in (1), while the 59 negative instances all exhibit main verb negation, as in (2):
(1) (1) _S1B_064_106> It is possible that they may have seen my advertisement as well
(2) (2) _S1A_069_271> Uhm so I think he may not have the confidence to go ahead as it were
2.2. Dynamic modality: possibility
In the corpus under scrutiny, 'dynamic modality' is expressed in 300 cases (25% of the total), which are particularly concentrated in formal, technical, scientific contexts, such as the following:
· · academic writing - natural sciences
· · non-academic writing - natural sciences
· · non-academic writing - technology
Unlike instances of epistemic possibility, where the speaker/writer puts forward his/her point of view quite overtly, in the occurrences of dynamic possibility, the speaker/writer is merely relating a state of fact, made possible/impossible by external circumstances, as in (3):
(3) (3) _W2D_014_18> The dimensions of the antenna are directly related to the wavelength of the signal you intend to receive and the elements of the antenna are set at right-angles to the transmitter, and may be aligned vertically or, more commonly, horizontally, to match the polarity of the transmitted signal.
Only 7 occurrences of MAY NOT conveying dynamic possibility have been recorded in the corpus, 6 of which exhibit main verb negation, as in (4), while only 1 is an instance of modal verb negation, namely (5):
(4) (4) _S2A_058_139> And this collagen may or may not have fibroblasts in it
(5) (5) _W2A_012_39> This is not to say that large, well-organized, and long-established long established religions may not be monolithic - history clearly shows us that they can be, but it is to suggest that the more established a religion is in a pluralistic society, the more 'internal pluralism' it is likely to display.
2.3. Deontic modality
The 173 instances of deontic modality (14% of the total) occur most often in 'parliamentary debates', 'legal cross-examinations', 'business letters', and 'administrative/regulatory writing'. Four different speech acts are encoded in these deontic values: 'regulation', 'request', 'permission', and 'wish':
· · regulation:
(6) (6) _W2D_006_79> NO BOOK OR OTHER PROPERTY OF THE BRITISH LIBRARY AND NO MATERIAL TEMPORARILY IN THE CARE OF THE BRITISH LIBRARY MAY BE REMOVED FROM THE ROOM IN WHICH IT WAS ISSUED.
· · request:
(7) (7) _S1B_053_2> May I ask her whether she thinks, that the eleven are not now both isolated and intransigent in relation to, agricultural policy
· · permission:
(8) (8) _S1B_062_2> If you wish to be seated you may with My Lord 's permission
· · wish:
(9) (9) _S2B_041_51> My lords and members of the House of Commons I pray that the blessing of Almighty God may rest upon your counsels
The present paper has focussed on the modal verb MAY as it occurs in the British Component of the International Corpus of English. However, an exhaustive picture of this verb can only be drawn if we do not lose sight of the wider canvas of the whole modal system, which will involve studying other verbal and non-verbal realizations of the same semantic and pragmatic values that have been discussed for MAY. Moreover, a consistent study of any modal verb is bound to tackle also the knotty issue of the different types of modality, since the epistemic and deontic categories cover only part of the semantic realm, while a third type of modality, generally labelled with the term 'dynamic', is needed to qualify the linguistic realizations of a number of modal elements. Hence, rather than being intended as a self-contained study, the present analysis should also be considered as a means to further qualify the field and the boundaries of the types of modality, including dynamic modality itself, which has often been excluded from the semantic realm pertaining to MAY.
Aarts, B. and Meyer, C.F. (eds.) (1995) The Verb in Contemporary English. Cambridge University Press, Cambridge.
Aijmer, K. (1997) Epistemic Modality as a Discourse Phenomenon - a Swedish-English Cross-Language Perspective. In Fries, U., Müller, V. and Schneider, P. (eds.), pp.215-226.
Biber, D., Johansson, S., Leech, G., Conrad, S. and Finegan, E. (1999) Longman Grammar of Spoken and Written English. Longman, London.
Bybee, J. and Fleischmann, S. (eds.) (1994) Modality in Grammar and Discourse. Benjamins, Amsterdam.
Coates, J. (1995) The Expression of Root and Epistemic Possibility in English. In Aarts, B. and Meyer, C. F. (eds.), pp.145-156.
Facchinetti, R. (2000) Be able to in Present-day British English. In Mair, C. and Hundt, M. (eds.), pp. 117-130.
Fries, U., Müller, V. and Schneider, P. (eds.) (1997) From Ãelfric to the New York Times. Studies in English Corpus Linguistics. Rodopi, Amsterdam.
Greenbaum, S. (ed.) (1996) Comparing English Worldwide: The International Corpus of English. Clarendon Press, Oxford.
Groefsema, M. (1995) Can, May, Must and Should: A Relevance Theoretic Account, Journal of Linguistics 31: 53-79.
Kirk, J. (ed.) (2000) Corpora Galore. Analyses and Techniques in Describing English. Papers from the 19th International Conference on English Language Research on Computerised Corpora (ICAME1998). Rodopi, Amsterdam.
Klinge, A. (1993) The English Modal Auxiliaries: From Lexical Semantics to Utterance Interpretation, Journal of Linguistics 29: 315-357.
Krug, M. (2000) Emerging English Modals: A Corpus-Based Study of Grammaticalization. [Topics in English Linguistics]. Mouton de Gruyter, Berlin.
Leech, G. (forthcoming) Diachronic Linguistics across a Generation Gap: From the 1960s to the 1990s. Paper presented at Grammar and Lexis, University College London, 21st July 2000.
Lichtenberk, F. (1994) Apprehensional Epistemics In Bybee, J. and Fleischmann, S. (eds.), pp. 293-327.
Mair, C. and Marianne, H. (eds.) (2000) Corpus Linguistics and Linguistic Theory. Papers from the Twentieth International Conference on English Language Research on Computerized Corpora (ICAME 20) Freiburg im Breisgau 1999. Rodopi, Amsterdam.
Myhill, J. (forthcoming) Themes in the Historical Development of American English Modals: From Social to Individual to Impersonal. Paper presented at 11th International Conference on English Historical Linguistics, Santiago de Compostela, 7-11 September 2000.
Palmer, F. R. (1990) Modality and the English Modals. (2nd ed.) Longman, London.
Palmer, F. R. (1994) Negation and the Modals of Possibility and Necessity. In Bybee, J. and Fleischmann, S. (eds.), pp. 453-471.
Palmer, F. R. (forthcoming) Negation and the Modal Verbs in English. Paper presented at Grammar and Lexis, University College London, 21st July 2000.
Papafragou, A. (2000a) Modality: Issues in the Semantics-Pragmatics Interface. Elsevier, Amsterdam.
Papafragou, A. (2000b) On Speech-Act Modality, Journal of Pragmatics 32: 519-538.
Vihla, M. (2000) Epistemic Possibility: A Study Based on a Medical Corpus. In Kirk, J. (ed.), pp.209-224.
Winford, D. (2000) Irrealis in Sranan: Mood and Modality in a Radical Creole, Journal of Pidgin and Creole Languages 15(1): 63-126.
Gaëtanelle Gilquin (Université catholique de Louvain, Belgium)
Although there is a plethora of studies from various linguistic trends devoted to causative constructions (e.g. Shibatani 1975, 1976, Ritter & Rosen 1993, Song 1996), there is scope for further research in the field. This study has two distinctive features. First, it does not deal with ‘the causative construction’ in general, but exclusively focuses on four causative verbs, viz. cause, get, have and make, the most frequent periphrastic causatives with no or little semantic content on their own, apart from the causative meaning of ‘bringing about’ (unlike for instance force, which, besides causation, also clearly expresses coercion). Although these verbs are dealt with in most grammars, no satisfactory account is given of the circumstances in which each of them should be used, nor the consequences of the use of a particular type of complement (infinitive, past participle or present participle). Second the study is based on corpus data, so that it should give a better idea of how causative verbs behave in authentic present-day English (precise meaning, frequency, diatypic variation, combinatorial properties, etc.).
This presentation will fall into two parts. First, I will show how causative constructions such as John made her laugh or I had my watch repaired can be retrieved from corpora (semi-) automatically. More specifically, I will compare the results achieved with a concordancer like XKwic, a piece of software developed at the University of Stuttgart which can carry out highly refined and specialised linguistic searches, and with ICECUP 3.0, the program designed to query the International Corpus of English (ICE) and working on the basis of ‘Fuzzy Tree Fragments’ representing the grammatical structure of sentences. This comparison will highlight the fact that, although ICECUP has higher precision and recall rates, as long as it cannot be used in conjunction with other (larger) corpora, a more ‘classical’ concordancer will be needed to thoroughly investigate relatively rare phenomena such as periphrastic causative constructions.
Secondly, I will present the preliminary results reached on the basis of ICE-GB (1,000,000-word corpus), the British component of ICE. Following the functional ‘one meaning, one form’ principle, I put forward the hypothesis that there must be differences between the four causative verbs cause, get, have and make. In order to test this hypothesis, the causative sentences retrieved were examined both quantitatively and qualitatively with respect to a number of syntactic, stylistic and semantic parameters. The syntactic survey focused on the types of structures that are available for each causative (bare infinitive or to-infinitive, present participle, main clause or subclause passivization). From a stylistic point of view, I investigated whether the four verbs and their non-finite complements were stylistically differentiated by comparing their frequencies in speech and writing, as well as in the different genres of ICE (e.g. novels/stories, business letters, face-to-face conversations, etc.). Semantically speaking, finally, I followed Fillmore and his theory of Frame Semantics (cf. the FrameNet Project) in viewing causative constructions as made up of three ‘Frame Elements’, viz.
The explosion caused the temperature to rise.
Cause Affected Effect
Each Frame Element can be described in terms of various features, such as animacy of the Cause and the Affected, volitionality of the Effect, or degree of coercion involved. The semantic study is complemented with a collocational analysis, whose aim is to determine the preferential lexical company kept by each causative verb. It should be emphasised, however, that these results are based on a relatively small number of instances (40 constructions with cause, 101 with get, 77 with have and 150 with make) and therefore need to be substantiated by further and more extended research.
I wish to acknowledge the support of the Belgian National Fund for Scientific Research.
Gilquin, G. (1999) Causative ‘make’: A corpus-based study. Unpublished M.A. dissertation, Université catholique de Louvain.
Gilquin, G. (2000) Periphrastic causative verbs ‘get’ and ‘have’. Towards a systematic description. Unpublished M.A. dissertation, Lancaster University.
Shibatani, M. (ed.) (1976) The Grammar of Causative Constructions. Syntax & Semantics 6. Academic Press, New York/San Francisco/London.
Song, J.J. (1996) Causatives and Causation. A Universal-Typological Perspective. Longman, London/New York.
Xkwic, IMS, Stuttgart (http://www.ims.uni-stuttgart.de/projekte/CorpusWorkbench)
This paper aims to analyse the use of SHALL and WILL for the formation of the first person subject future tense in a corpus of Early Modern English texts. The period taken into consideration is 1640-1710 and the texts analysed are those included in the third section of the Early Modern English part of The Helsinki Corpus of English Texts.
The use of the future tense formed with SHALL and WILL has been the object not only of various previous analyses, but also of specific rules pointed out in several of the grammar books published in the seventeenth century. As regards the former, one of the first studies carried out on the subject is Fries (1925), whose investigation is based on a survey of the usage of SHALL and WILL in fifty English dramas from 1560 to1915 (two dramas of roughly the same date were selected for approximately every decade in that period). In examining these texts, Fries divided the instances into three groups: (1) WILL and SHALL in independent declarative statements; (2) WILL and SHALL in questions; (3) WILL and SHALL in subordinate clauses. As regards the first group, with first person subjects WILL has been found to be more frequently used than SHALL (with average percentages of respectively 80% vs 20% in the period taken into consideration here). On the other hand, in direct questions SHALL overwhelmingly predominates (97% vs 3%); even in the few instances in which WILL occurs in the first person, the majority of cases consists of ‘echo-questions’, in which WILL repeats the use of the same modal auxiliary in the previous sentence. As regards subordinate clauses, the data are fairly well-balanced, with a slight majority of shall-forms (53.4% vs 46.6%).
In her analysis of a corpus of texts taken from a wider section of the Helsinki Corpus than ours, Merja Kytö (1991) finds an initial increase in the use of WILL with first person subjects, reaching a peak from the 1570s to the 1640s; this increase in the use of WILL is particularly noticeable in colloquial language (e.g. private letters) and speech-based texts (e.g. sermons and trial proceedings). Later the use of WILL decreases, probably owing to the regulating influence of grammarians, who started advocating SHALL in first person and WILL in second and third person uses. As regards the period taken into consideration here, the use of the two modal auxiliaries with first person subjects is quite well-balanced, with a slight majority for SHALL (51.7% vs 48.3%); WILL, however, occurs more frequently with dynamic uses of the main verb (in 71% of all cases), while SHALL is the auxiliary favouring stative uses (in 66% of all cases). In direct questions, instead, first person SHALL is dominant (100% of all cases).
The results of these studies will be compared to the rules of usage pointed out in a few 17th-century grammars, namely Wallis (1653), Cooper (1685), Miege (1688) and Aickin (1693). The analysis of these grammars will help to outline the main uses of the two auxiliaries in future contexts identified by early English grammarians, while a careful analysis of the two previous studies will provide important points of comparison with the data found in our research.
Our analysis will focus on the uses of SHALL and WILL with first person subjects both in interrogative and non-interrogative sentences, and will examine their occurrences (both from a quantative and a qualitative point of view) in different text types and for the performance of various pragmatic functions (e.g. prediction, intention, promise, proposal).
The analysis of the corpus will show that the rules laid down by the grammars of the period taken into consideration oversimplify the range of uses of the two modal auxiliaries; indeed, first person future expressions with SHALL do not merely denote prediction or declaration, nor do those with WILL only express promise, intention or resolution.
SHALL is used in interrogative sentences to express various pragmatic functions, such as asking for the addressee's opinion, giving suggestions, requesting advice, and inquiring about the addressee's wishes. In addition to these uses, questions starting with shall-forms perform a predictive function; occasionally they are used as rhetorical devices to serve argumentative purposes. In non-interrogative sentences SHALL mainly indicates prediction; indeed, for the expression of this pragmatic use this modal auxiliary is employed in almost an exclusive way. Shall-forms are also used to express intention, representing an alternative to the use of will-forms; however, the quantitative difference between the two modals is relevant.
WILL is very seldom used to express prediction in first person subject statements; as this use has only been found in one text, its occurrence may be due to the author's idiosyncratic behaviour rather than general usage. The most widespread pragmatic function performed by first person subject will-forms is that of intention; for the expression of this function WILL has a much higher percentage than SHALL. In particular, a comparison in the use of the two modal auxiliaries in homogeneous contexts points to the adoption of WILL where a more marked degree of intentionality is to be denoted.
First person subject will-forms are also used to express promise; indeed, they represent the main form of expression of this pragmatic function. Some occurrences of first person WILL statements represent instances of proposals, a speech act which relies exclusively on this modal. For the expression of this speech act SHALL is used in questions starting with Shall we, but no instances of shall-forms have been found in non-interrogative sentences with this purpose.
Aickin, J. (1693) The English Grammar, London.
Cooper, C. (1685) Grammatica Linguae Anglicanae, London.
Fries, C. C. (1925) The Periphrastic Future with Shall and Will in Modern English. In Publications of the Modern Language Association of America, XL, pp. 963-1024.
Kytö, M. (1991) Variation and Diachrony, with Early American English in Focus. Peter Lang, Frankfurt am Main.
Miege, G. (1688) The English Grammar, London.
Wallis, J. (1653) Grammatica Linguae Anglicanae, London.
1.1 Introduction. The problem of idiomaticity has been traditionally perceived as one of the greatest challenges in the field of linguistics and foreign language teaching, and is particularly true of analytical languages like English which tends to 'isolate' its units not only structurally but also semantically. In spite of a wide range of construction types of idioms represented in the English language, the term 'idiom' has been largely applied without distinction to pattern. It is mostly viewed as a semantic matter manifested in much the same way in expressions of different structural types. What makes a particular expression idiomatic is its semantic globality which is difficult to interpret in terms of the meanings of its constituent words. Idioms vary from opaque to relatively transparent, they are not divided as a small water-tight category being related to non-idioms along a scale or continuum (A.P. Cowie The Treatment of Collocations and Idioms in Learner's Dictionary).
The view taken in the present paper is that in text reality we are faced with the rich diversity of established phrases which present both lexical and syntactic units. They fall under the notion of phraseology because "one is first struck by the fixity and regularity of phrases, then by their flexibility and variability" (John Sinclair Corpus Concordance Collocation, OUP, 1991, p.104). Some of such phrases or word-combinations are characteristic not so much of the system of language (lexical units) but of speech tendencies accounting for the 'naturalness', recurrence, and utility of particular items (syntactic units).
The dichotomic division of all word-combinations into 'free' and idiomatic is seldom justified. This relationship is observable as a gradation in degree of idiomaticity: every time we concern ourselves with the particular sequence of words it is never a static assignment of this or that phrase to either part of the above dichotomy but the dynamic unity of colligation and collocation. Phraseology covers both idiomatic units proper and those which without being truly global semantically still remain 'fixed', 'set' or recurrent reflecting the typical lexical and syntactic choices. Hence the division into idiomatic and non-idiomatic phraseology.
1.2. The aim of the present paper is to try and approach the problem of idiomaticity ('choice and arrangement of words') through a contrastive corpus-based analysis of most typical phraseological choices made by Russian learners of English as compared to those of native speakers.
To cope with the large number and great diversity of word-combinations the following semantic criteria (categories) have been elaborated:
· · connotativeness - non-connotativeness - accounting for the word-combinations performing the emotive function as distinct from the referential one;
· · cliché-ed expression - revealing the opposition of set expressions and those created anew for the particular situation;
· · idiomaticity - showing a chain-like gradation in degree of 'opaqueness' from idioms to non-idioms;
· · conceptual integrity - considering the referential basis of word-combinations;
· · cultural and sociolinguistic determination - addressing the extralinguistic or cognitive nature of linguistic items.
A contrastive analysis of idiomatic phraseology used by native and non-native (Russian) speakers of English was carried out on the basis of three components of the International Corpus of Learner English (ICLE): British Component (95,695 words); American Component (153,348 words) and Russian Component (228,846 words).
In order to investigate the peculiarities of various types of idiomatic expressions realized in speech production we focus on the Adjective + Noun combinations which present the classic example of the word combination and display the basic properties of this kind of construction. From the chosen corpora we have retrieved the following number of word-combinations:
Component of ICLE
Total number of Adj + N
2.1. A contrastive analysis of the Russian and the Native Speaker (British and American) components of the International Corpus of Learner English (ICLE) has yielded the following results:
1) 1) High frequency Adj + N word-combinations with the frequency band of 111 – 10, on the one hand, reflect thematic peculiarities of the corpora which consist predominantly of student essays. On the other hand, such expressions represent not only the subject matter of each corpus but also the sociolinguistic features characteristic of a particular speech community. Thus, the most frequent items in the American Component, for example, are United States (111 occurrences); ethnic American (76); American literature (63); public schools (57) and capital punishment (50) while in the British Component the top five word-combinations include: bad faith (54); prime minister (53); single Europe (44); European community (35) and philosophical optimism (31).
2) 2) The data have shown a much stronger tendency on the part of native speakers to use idiomatic (for example, at the same time, a great deal, common sense) and cliché-ed expressions (such as natural disasters, free trade, racial prejudice) as compared to the Russian students of English:
Component of ICLE
(% of occurrences)
(% of occurrences)
3) 3) British, American and Russian components of ICLE include a certain amount of word-combinations which cannot be characterized either as idiomatic or cliché-ed. The difference between the Native and Non-native speaker corpora in this case lies in the semantic and functional peculiarities of the word-combinations under consideration. In the British and American components such expressions reflect occasional and/or context-bound uses of words, such as overnight visitation, symbolic interactions, exigent circumstances, etc. In the Russian Learner Corpus, however, these are the word-combinations which in most cases, though conceptually integral and understandable, are not found in common use and therefore can be described as collocational errors or literal translation from the mother tongue. For example, good purposes, broad education, material things, good sides, pure feelings, etc. The availability of these data reflecting different tendencies of native and non-native speakers to use word-combinations with various degrees of idiomaticity appears to be quite instrumental in the area of English language teaching.
Sebastian Hoffmann (University of Zurich, Switzerland)
Complex prepositions are traditionally considered to be fixed units that are indivisible both in terms of syntax and in terms of meaning. Thus, they form the head of a prepositional phrase (PP) and need not be analysed into smaller constituents (e.g. PP -> P + NP + P (+ NP)). Typical examples are shown in (1) and (2):
(1) They went swimming in spite of the rain.
(2) In breach of the protocol, he left the banquet early.
However, most complex prepositions allow the insertion of further elements such as premodifying adjectives or determiners (e.g. in hot pursuit of X). Quirk et al's "scale of cohesion" (1985:671f.) takes account of this fact and lists nine possible types of internal variability, with the prototypical preposition in spite of at the one extreme - allowing none of the variations - and the syntactically much less restricted on the shelf by at the other end of the scale (and allowing all 9 types of variation). While this clearly offers a convincing descriptive approach to complex prepositions, it does not address the question of how the concept of a syntactically indivisible unit can be reconciled with clear indications of an internal grammatical structure.
This point was taken up by Seppänen et al. (1994), who applied four standard constituency tests to the constructions in question: coordination, interpolation, fronting and ellipsis. Consider examples (3) - (6):
(3) It will undoubtedly need further refinement and modification in the light of consultations and of experience. BNC:ANS:349
(4) In spite, therefore, of the transformations, basic comfort was lacking. BNC:ANR:478
(5) The debate centred around the issue of state control of education, of which Williams was in favour , and in particular the Aristotelian view on the issue. BNC:GXG:2158
(6) A: In the light of what you've said, I agree to the changes.
B: Of what I've said! Don't put the onus on me! Seppänen et al. (1994: 21)
Each of the sentences in (3) to (6) suggests that there is in fact a constituent boundary between the noun and the second preposition in complex prepositions. After presenting a whole range of examples for both 2-word and 3-word complex prepositions, Seppänen et al. conclude: "Introduced into the grammar on the basis of an untenable analysis, the class of complex prepositions as defined by Quirk et al. is empty, and the term itself is thus not helpful in the description of English" (1994: 25).
My paper is at least in part intended as an extension of Seppänen et al.'s work. The authors do not offer corpus-based evidence for their claims but rely on native-speakers as informants (Joe Trotta, personal communication). I do, however, believe that corpus data can offer some new insights that may in fact lead to a somewhat different evaluation of the argumentation presented. For this purpose, I will provide data from the 100-million word British National Corpus (BNC).
My study is based on the 25 most frequent complex prepositions in the BNC. Together, they account for just over 60,000 tokens in the whole corpus and thus cover about 60 per cent of all constructions commonly held to be complex prepositions. I will be concentrating on two of the four constituency tests discussed in Seppänen et al., namely coordination and interpolation.
If two strings can be coordinated, they must be constituents, and must normally be identical functionally and usually even categorically. (Seppänen et al. 1994: 13)
The following three variants of coordination with complex prepositions can be found:
(7a) In terms of household income and the number of working hours per week ...
(7b) In terms of household income and in terms of the number of working hours per week ...
(7c) In terms of household income and of the number of working hours per week ...
The data gathered from the BNC shows that constructions (7a) and (7b) account for the overwhelming majority of instances. Although this does not per se invalidate the argumentation presented in Seppänen et al., it nevertheless suggests that when given a choice, speakers will treat complex prepositions as a unit1. I will also draw on further data from a subclass of coordination - correlative coordination - in order to present a more complete picture of the choices made by users of the language.
When elements are added to a structure, the new elements may be inserted at some of the constituent boundaries of the clause, with heavy restrictions depending on the particular case in question, but such interruption is totally impossible with items which, in spite of rich internal structure, function as single units with no syntactic constituent boundaries between them. (Seppänen et al. 1994: 13)
The data for interpolation shows a more evenly distributed picture for the different choices available to speakers of English and thus offers less conclusive evidence for any preferences taken. I will discuss interpolation as an indicator of constituent boundaries and show with the help of corpus data that this test is less relevant for the present context than is suggested by Seppänen et al.
Following this general discussion of constituency tests, I will show that only a few types of complex prepositions (most importantly in terms of) make up the bulk of the sentences supportive of Seppänen et al.'s argumentation and will discuss the extent to which this, too, necessitates a re-evaluation of their approach.
Finally, I also hope to present the first results of a collocation-based approach to the retrieval of complex prepositions, extending the common formulae for the calculation of collocational strength from simple node - collocate pairs to constructions with two or more collocates (e.g. the node in followed by a noun which itself is followed by a preposition within a certain number of words, such as in (...) pursuit (...) of). The data will be used as a testbed for the claims presented in the literature.
Seppänen, A., Bowen, R. and Trotta, J. (1994) On the So-called Complex Prepositions, Studia Anglica Posnaniensia 29: 3-29.
This is a work-in-progress report of a project launched in 1999 for compiling a one-million word speech corpus of Japanese learners of English. One of the main characteristics of the TAO Corpus is that the corpus data is entirely based upon the audio-recordings of an English oral proficiency interview test called the Standard Speaking Test (SST). Based on the well-known ACTFL OPI (Oral Proficiency Interview), this 15-minute interview test consists of several tasks including picture description, role-playing, and story telling. One of the unique features of the corpus is that each speaker's data includes his/her proficiency profile based on the SST evaluation schemes, SST Level 1 to 9.
We are planning to make this corpus open to the public so that teachers and researchers of many kinds can use it for their own research interests: SLA, syllabus design, or natural language processing (NLP), etc. Our purpose, as an NLP research group, is to create a computerized pedagogical tool which can process inputs containing learners’ errors. We will do this mainly by analysing learners’ errors from each proficiency level with error tagging, then by constructing a model of learner English across different proficiency levels.
In this report, we will introduce this new Japanese CLE project, showing its data collection procedure, the original tool for transcription, annotation schemes and also explaining how to apply this data for the development of a pedagogical tool.
Randall L. Jones (Brigham Young University, USA)
A language corpus can be a useful tool for compiling information about word frequency. Most software that is used for corpus analysis has tools that make it rather simple to generate information about both relative and absolute frequency of words in the corpus. Such information is useful, inter alia, for general lexical research as well as for second language learning and teaching. It is assumed that in the process of learning a new language frequency of usage is an important criterion in the sequencing of vocabulary, i.e. the words used most frequently by native speakers should be those that are learned first. Vocabulary lists, instructional material, and even frequency dictionaries can thus be based on this information.
Interest in German word frequency dates back at least to the 19th Century, when F. W. Kaeding published his monumental Häufigkeitswörterbuch der deutschen Sprache. Festgestellt durch einen Arbeitsausschuß der deutschen Stenographiesysteme in 1898. The work is an object of wonder for several reasons, not the least of which is the fact that it is based on a 10 million word corpus taken from 94 separate sources before the advent of the computer. The nearly 700 word book which is full of lists and charts was published by Kaeding himself. In 1928 Professor B.Q. Morgan of the University of Wisconsin published a re-worked version of Kaeding’s list intended for the teaching of German.
There have been numerous frequency lists of German produced in the computer era, both published and unpublished. One of the most useful was that of Professor J. Alan Pfeffer of the University of Buffalo in 1964. It is based on his Grunddeutsch material, which is a spoken corpus of German compiled in 1960 and consisting of ca. 700,000 running words. It was the basis for vocabulary selection in many of the elementary German text books published in the United States during the 1960s and 70s. Pfeffer’s list is the result of careful analysis and sorting. It contains only headwords or lemmata, and not the numerous word forms.
Although the computer is a marvelous tool for generating word frequency information, the raw frequency data generated from a corpus is only the first step in the process of compiling a useful pedagogical tool for vocabulary learning. Many of the words must first be lemmatized and disambiguated. As is well known, lemmatization means collecting all inflected forms of a word into a single headword or lemma, e.g. German bin, bist, ist, etc. under the headword sein (to be). The process of disambiguation separates homographs into their respective meanings, e.g. sein (verb form and possessive pronoun), wohl (particle and adverb), and heißen (verb form and adjective).
The processes of lemmatization and disambiguation in German are not as straightforward as may at first appear, especially if the corpus is not tagged. Many inflected forms of verbs, adjectives and nouns can belong to one of several parts of speech, thus a good deal of human intervention is necessary. For example, if one disregards capitalization, the German form liebe can be a noun, a verb, or an adjective. It becomes necessary to examine each form to see how to assign it to its proper lemma. Even more difficult is the sorting out of homographs, as it is not always obvious how many separate words a single form may represent, even within the same word class.
This paper will address these and other issues that are currently confronting the author in the construction of a pedagogically oriented frequency dictionary of modern German. It is based on his existing BYU Corpus of Spoken German as well as a written German corpus which is now in the process of being compiled. The written corpus comprises three million words and consists of a collection of printed material containing samples in the three categories of journalism, literature, and non-fiction prose, all published since 1990. One of the challenges of the project is to balance the two corpora (700,000 vs. 3,000,000 words) as well as control the range of the entries so that a word is not misrepresented because of a preponderance of examples in a small number of texts. The final product will be an annotated frequency dictionary of German that will contain the most frequent 3,000 words in the corpus together with brief examples of their usage. The examples will be authentic German based on the corpus material.
Jones, R. L. (1997) Creating and Using a Corpus of Spoken German. In Wichmann, A., et al. (eds) Teaching and Language Corpora. London: Longman, pp. 146-56
Kaeding, F. W. (1898) Häufigkeitswörterbuch der deutschen Sprache. Festgestellt durch einen Arbeitsausschuß der deutschen Stenographiesysteme. Selbstverlag des Herausgebers, Steglitz bei Berlin.
Morgan, B. Q. (1928) German Frequency Word Book. Based on KAEDING’s Häufigkeitswörterbuch der deutschen Sprache. McMillan, New York.
Pfeffer, J. A. (1964) Basic (Spoken) German Word List. Grundstufe. Prentice-Hall, Englewood Cliffs
Przemyslaw Kaszubski (Adam Mickiewicz University, Poland)
I would like to report in this paper on some technical problems encountered in my recently completed PhD work (http://main.amu.edu.pl/~przemka/rsearch.html). In the project, I set out to examine the quantitative correlation between several bands of proficiency in written English and relative frequency of use of idiomatic and non-idiomatic expressions containing high-frequency primary verbs. Findings for two such lemmas - GIVE and TAKE - will be cited in the presentation to illustrate my points. The corpus network applied consisted of small, argumentative/ expository collections (partly pooled from the International Corpus of Learner English resource), featuring advanced and intermediate EFL learner varieties, native English learner varieties, native English expert writing, and contrastive native-tongue material (Polish).
Most learner-corpora projects tend to rely on small, 'intimately known' text resources which are carefully read, edited, annotated, etc. The necessity for a close relation with interlanguage data is a methodological strength but also a serious drawback since many less frequently distributed phenomena (such as specific collocations and idioms) cannot then be studied convincingly with statistics. In my view, two major kinds of factors, at least with respect to lexical studies, contribute to this bottleneck: language- based and technical factors.
Language-based difficulties relate to the quality of language in the investigated text corpora, often reflecting the principles of corpus design and corpus compilation. Besides the statistical question of representativeness, quantitative learner language research typically relies on inter-corpus comparability, making it necessary to control important variables, such as language task type; author age, proficiency and demographic characteristics; text length; etc. Unfortunately, the more corpora are involved in a project, the less easy it is to exercise control over their linguistic and textual homogeneity, unless one chooses to exploit self- compiled data only, which is usually not the case. One inescapable problem with interlanguage corpora is their effective stratification according to proficiency. With this, rough statistical measures, such as standardised type-token ratios, may prove useful, as I will attempt to illustrate.
Many of the language-based difficulties limit possibilities for computerised research. For instance, although advanced students' writing has been found not to upset the effectiveness of part-of-speech taggers (Meunier 1998: 21), comparison with lower-level EFL texts, often fraught with misspellings, syntactic errors and even punctuation irregularities, may be seriously hampered. In fact, the error-margin left by a speech tagger may adversely affect even native English findings, especially if the type of mistagging is consistent and goes unnoticed. Distinguishing between verbal and non-verbal (nominal, adjectival, auxiliary etc.) occurrences of English verbs can be helped but by no means resolved by taggers, as the example of the passive vs. pseudo-passive vs semi-passive differentiation can show only too well. Unfortunately, making such fine distinctions is often essential in pedagogically oriented studies.
Collocational and phraseological studies depend on other kinds of disambiguation as well. Correct word sense disambiguation, although being developed e.g. through the Senseval Project, is far from at an implementable stage. A different, Firthian way to finding out about word meaning would be through building information about its collocational network. However, automatic collocation extraction by means of recurrent word strings or even collocation statistics (MI, z-score, e.g. in WordSmith Tools or TACT 2.1) often produces too much noise in the outcome (low precision), and is not very efficient for small corpora (low recall). Another disadvantage of cooccurrence statistics is that often input cannot be manipulated in the way the researcher wishes, and the values are calculated from all rather than selected occurrences of wordforms or lemmas (a notable exception is Oliver Mason’s QWICK).
In the end, an applied researcher studying multi-word units in learner data is forced to find his own solutions to faster annotation and retrieval of data. I would like to offer a few such suggestions, based on my recent empirical work and the following resources: a POS tagger/lemmatiser (TOSCA-ICLE Tagging Unit); corpus analysis tools (WordSmith Tools, TACT), a text editor (Word97), a spreadsheet program (MS EXCEL 5.0) and GNU UNIX-based text-editing programs ported to DOS (awk, sed, cut etc.). At the end of the discussion, I will call for a more universal concordance editing/extraction tool which, I believe, could serve to ease many of the presented limitations in learner-corpus research.
The quick-and-dirty solutions may comprise many levels of disambiguation, including:
a) the use of spell-checking facilities and POS-tagging software for flagging and fixing undesirable spelling, syntactic and other surface mistakes;
b) the use of GNU-UNIX text analysis tools to target and edit unresolved ambiguities and heuristic tag assignments, and thereby to preclude skewed statistics;
c) (less or more detailed) manual disambiguation of concordance lines with the WordSmith Tools concordance annotation facility;
d) (when collocations stats are planned): saving (roughly) disambiguated concordance lines in text format to enable the concordancer to re-read, and draw statistics for, the disambiguated set rather than all the occurrences of a given family of wordforms;
e) assessing disambiguated collocations / expressions and targeting statistically skewed types by measuring standard deviation across the corpora, establishing a cut-off point and double-checking for topic distribution in the textual data.
More enhancements could be added if necessary. It seems that linguistically reliable studies of much larger learner corpora could be possible if analysts had access to programs enabling enhanced on-line editing from a set of concordance lines. Such ‘power-concordancers’ should allow editing both the original text AND POS tags AND lemmatisation AND custom-made annotation. The results of completed disambiguation processes should then be conveniently saved in a text file (preferably pasted into the original corpus). They then could be called back into the concordancing module for further analysis, e.g. for conducting relevant statistical tests. Specifically for English phraseological studies, the enhanced editing-concording package could be equipped with an on-line dictionary of multi-word items to speed the process of disambiguation or inform / supplement the collocation statistics calculation.
The secret of successful learner corpus analysis is in making the indispensable stage of manual editing as fast as possible. This will not happen if there are no flexible and powerful tools available that ordinary applied researchers, rather than dedicated corpus analysts or programmers, can find accessible and useful.
Bernhard Kettemann (University of Graz, Austria)
This paper examines the usefulness of the Oxford English Dictionary (the OED) in relation to the British National Corpus (the BNC). I will try to find out how they complement each other in a semantic analysis.
The semantic analysis is concerned with the different meanings of the morpheme eco. Eco is not only one of the most fashionable prefixes in English these days, in times of the Kyoto protocol renunciation, BSE and foot and mouth disease, everything eco seems also politically relevant.
1 Additive and contrastive analyses
Basically, we can take two routes if we have two sources of data, we can do an additive analysis or a contrastive analysis.
Additive analysis means that we simply add up the data from the two sources so that we end up with a superordinate pool of data which we then analyse.
Contrastive analysis means that we take both sources of data and analyse them separately and then contrast the findings, taking into account the different natures of the two sources.
It seems to make more sense to embark on a contrastive analysis of eco in the OED and the BNC. This, however, first requires a discussion of the actual differences between these two kinds of sources.
2 The BNC and the OED – two types of data
The most interesting difference between corpora and dictionaries is that the eco-words in the latter are filtered, which means that not all words that have been found are entered. The selection is based on the criterion of institutionalization. Institutionalization means that a lexeme has a widely accepted intersubjective status.
The BNC cannot distinguish between institutionalized and less widely accepted lexemes. A dictionary, however, fulfils the purpose of representing institutionalized words by being the product of a number of linguistically-trained or at least highly language-conscious lexicographers filtering. It will therefore be interesting to see whether there are crucial differences between the eco-words found in the BNC and those in the OED.
3 The semantics of eco
The range of meanings that eco- assumes as a prefix is directly linked to the meanings of the neo-classical compound ecology and its derivates, particularly the adjective ecological. This means that there are six meanings of eco:
1. 1. pertaining to the study of the interactive relations between organisms
2. 2. pertaining to a more integrative study in any field
3. 3. pertaining to the (balanced) interaction between organisms / environment
4. 4. pertaining to the (balanced) interaction of entities within any field
5. 5. pertaining to the ecological movement
6. 6. environmentally friendly
4 Pertaining to the (balanced) interaction between organisms / environment
Table 1 lists all eco-words in the BNC and the OED where the prefix eco has the meaning 'pertaining to the interaction between organisms/environment' and where the base is itself a word of Greek origin. Those words that occur in both sources have been marked in bold print.
Table 1: Eco-words with Greek base and meaning 'pertaining to the (balanced) interaction of organisms/environment'
The facts that three of the ten most frequent eco-words in the BNC are contained in this group (ecosystem¸ ecotype, ecosphere) and that the majority of the terms also occur in both sources leads to the assumption that words of this pattern are relatively easily institutionalized. The reasons for this may be that the Greek base may lead to a feeling that the two parts belong closer together than in cases where the base is of a non-classical nature. Words of this sort further are typical of the socially revered scientific register.
In words such as eco-friendly, eco-sound(ly), eco-ok, eco-sensitive, eco-conscious, and eco-aware, eco has the same function, being added to a non-classical base meaning either ‘beneficial to’ or ‘conscious of’. Though according to the examples in the BNC, this seems a common process, these words lack the institutionalized appearance of the examples with the Greek / Latin bases as none of them occurs in the OED.
5 Environmentally friendly
Table 3 lists lexemes found in the BNC where eco has the meaning 'environmentally-friendly'.
Table 2: Lexemes with eco in the sense of ‘environmentally friendly’ in the BNC
Eco in the meaning of ‘environmentally friendly’ is a highly productive prefix, which shows in the high type frequency and the low token frequencies. As not a single word from this group has, however, entered the OED, I assume that the ‘environmentally friendly’ function is less likely to result in words that are easily institutionalized. This might be a result of associations with the discourse of advertising (note that all words marked with an asterisk are in fact brandnames), which is so quick to renew itself that the life-cycle of a word is not long enough to receive the honour of institutionalization.
Eco with the meaning ‘pertaining to the ecological movement’ is added to bases that are connected with politics. All examples of this kind are listed in table 3.
Table 3: Lexemes with the meaning ‘pertaining to / associated with the ecological movement’ in the BNC and the OED.
Even though the function of eco described above does not occur very often in the OED, it is remarkable that it appears at all. Together with the high number of types in the BNC this suggests that ‘pertaining to the ecological movement’ is a major meaning of this prefix, primarily as a productive morpheme but also one that to a certain extent also leads to institutionalization.
The other three meanings 1., 2., and 4. are rarely used according to the evidence provided by the BNC and the OED.
Of the six possible meanings of eco, three are highly productive. Of these, eco in the sense of ‘pertaining to the environment’, particularly if added to a Greek or Latin base, is most likely to result in a lexeme that will be quickly established. This is a result of the formal kinship of eco and the base, and also of the fact that these new lexemes appear to be derived from a scientific register. Eco in the sense of ‘pertaining to the ecological movement’ may result in increasing lexicalization because it is often used by the political establishment to attach a derogatory label to the relatively new green movement. Eco meaning ‘environmentally friendly’ is least likely to be accepted very quickly, perhaps because of its association with the ever-changing discourse of advertising.
What conclusions can we draw from our research question. "The BNC and or versus the OED"? This paper has shown that their different functions offer complementary insights into the semantics of English. For a complete analysis we need both types of data. So the answer is: the BNC and the OED, not versus, but apart.
Armstrong, S. (ed.) (1994) Using Large Corpora. MIT Press, Cambridge.
Aston, G. and Burnard, L. (1998) The BNC Handbook. Edinburgh University Press, Edinburgh.
Barnbrook, G. (1996) Language and Computers. Edinburgh University Press, Edinburgh.
Barton, D. (1994) Literacy. An Introduction to the Ecology of Written Language. Blackwell, Oxford.
Bauer, L. (1983) English Word Formation, Cambridge University Press, Cambridge.
Biber, D. (1993), Using Register-diversified Corpora for General Language Studies. In Armstrong, S. (ed.) (1994), pp. 219-241.
Crystal, D. (1995) The Encyclopedia of the English Language. Cambridge University Press, Cambridge
Crystal, D. (1997, 2nd ed.) The Cambridge Encyclopedia of Language. Cambridge University Press Cambridge.
Fill, A. (1993) Ökolinguistik. Eine Einführung. Gunter Narr, Tübingen.
McEnery, T. and Wilson, A. (1996) Corpus Linguistics Edinburgh University Press, Edinburgh.
Murison-Bowie, S. (1993) MicroConcord Manual. An Introduction to the Practices and Principles of Concordancing in Language Teaching. Oxford University Press, Oxford.
Owen, D. (1980), What is Ecology? Oxford University Press, Oxford.
Prechtl, P. and Burkard, F-P. (eds) (1996) Metzler Philosophie Lexikon. Begriffe und Definitionen. Metzler, Stuttgart & Weimar.
Quirk, R. (1974) The Linguist and the English Language, London: Edward Arnold.
One of the characteristics of a natural language is that it is largely systematic. This is what makes it an efficient communicative tool. As syntacticians, particularly those of the generative persuasion, have shown us so eloquently, the systematicity of language allows us both to say or write and to understand sentences we have never heard or seen before. If the syntactic module of language is thus clearly systematic, the lexical one is less clearly so. Nevertheless, there are large areas in the lexicon where systematic tendencies, not to say rules, obtain. This is the case in word formation. For example, presented with adjectives like readable or circular the native speaker expects there to be the derived nouns readability and circularity, perhaps in addition to other nouns. Inversely, the nouns readability and circularity presuppose the existence of the adjectives readable and circular. If this systematic bond between adjective and derived noun should break - if, say, we were to find readability but not readable - there would be a gap, an empty place in the system.
It is the aim of the present paper to investigate the existence and nature of lexical gaps with the help of corpus material. In order to do so all the adjectives in the Cobuild Corpus ending in the suffixes -ar, -ary, -ent, -ible, -ish, -ive were recorded along with their frequencies in the Corpus. For each such adjective its nominal derivations, if any, were recorded, again with their frequencies. The derived nouns ended in -arity, -ence, -ency, -ibility, -ishness, -iveness and -ivity, less frequently in -ariness, -aryism, -arialism, -arianism, -ibleness and -ivism. Whenever a noun with one of those suffixes was found, it was recorded, whether or not the corresponding adjective was found. When later the two lists, one of adjectives and one of derived nouns, were matched, it frequently turned out that items in one list had no equivalents in the other. Not only were nouns corresponding to the adjectives missing, the opposite case also occurred, although less frequently, viz. de-adjectival nouns having no base adjectives. Generally speaking, such gaps are to be expected in a corpus of whatever size, since in the lower frequency registers it is largely a matter of chance whether a word occurs or not. Therefore the cases considered to be of interest were only the most frequent of the adjectives with no corresponding derived nouns, and similarly only the most frequent of the nouns with no corresponding base adjectives.
It should be pointed out that the term "lexical gap" is here used in an operational sense, in that potentialities are not taken into account. Any one of the adjectives selected as the basis of the investigation has a potential noun derived from it (any adjective in, say, -al has a potential nominal derivation in -ality, etc.); likewise, any one of the de-adjectival nouns has a potential adjective as its base (any noun in -ality has a potential adjectival base in -al , etc.). However, if such a potential word does not occur in the Corpus, it, or rather its non-occurrence, is considered a gap.
When the most obvious gaps had been sorted out in this way, it appeared that a number of factors were operative in the field, some acting on one particular type of adjective or noun and some acting on most or all. It also became apparent that some of them work in conjunction. When the operative factors, or some of them, had been identified, they could account for the great majority of the gaps. Nevertheless, a few mystifying gaps remain to be explained.
A few general conclusions are that many lexical gaps are formal rather than functional, that at least some parts of the lexicon have a high degree of systematicity, and, unsurprisingly, that the lexicon is a very flexible component of language.
Natalie Kübler (Université Denis-Diderot Paris 7, France)
This paper presents a methodological approach to the teaching of terminology and specialized translation that is completely corpus-based and corpus-driven. It takes into account the significant impact corpus linguistics has come to have on linguistic thinking and teaching over the last years in France especially. This approach is dedicated to linguists, terminologists, and translators who are discovering corpus linguistics.
The tools we use consist of a commercialized term extractor (Terminology Extractor), a home-made Web-based concordancer using perl-like regular expressions, and a frequencer using the same kind of regular expressions.
Several corpora have already been collected, but this is a constantly ongoing process. Our corpora are mainly in French and English. The following types are readily available:
- - comparable corpora in general English and French;
- - comparable in two different English and French LSPs: computing and digital cameras;
- - parallel (or translation corpus) from English into French in the same subject areas;
- - monolingual specialized corpora either in French or in English.
Among others, some corpora that are currently being collected are comparable and translation corpora in genetic engineering, with texts that have been written in French and translated into English.
Translating LSP text requires an in-depth terminological analysis, as existing glossaries are often incomplete and do not contain enough information about phraseology, collocations, and translations into the target language, especially in fast developing subject areas, such as technical ones.
Our methodological approach deals with several steps in which translation and term extraction are intimately linked. Separating translation and terminology in the domain of LSPs can lead to missing important linguistic information.
In terminology and translation, a very general query must be made; this leads to various paths that must be followed one after the other. The search is narrowed little by little, using a list of criteria that have been isolated by scrutinizing corpora.
The approach, which will be sequentially described, step by step, involves a constant coming and going between monolingual, bilingual, general language, and LSPs corpora.
The first step consists in collecting a list of potential terms in the source language, using the Terminology Extractor on the texts to be translated. It is then necessary to query the specialized corpus in the source language with the following aims:
- - understanding the meaning of the various terms;
- - spotting multi-word terms that have not been highlighted by Terminology Extractor;
- - extracting as much information as possible about syntactic information and semantic classes;
- - analysing the textual environment in which terms appear; this step is most useful when the translation into the target language is difficult to find;
- - beginning to seek working hypotheses about possible translations.
A monolingual corpus of general language serves the purpose of checking whether a term is specialized or not, i.e. the term can have syntactic structures that do not exist in the general language.
Parallel corpora are used to find possible translations in the target language. Not all translations can be found in this way. That is why the first step is most important: the context around a term can help to find a possible translation in the monolingual target corpus.
Another necessary step consists then in comparing possible translations in the target corpus with their equivalents in the source text.
The translation of “activate a link” is in French “activer un lien”. However, adopting a systematic approach, one discovers other verbs governing the term “link”, which , in these cases is not translated into “lien” in French:
Any other link comes up n’importe quelle liaison démarre
After the link comes up une fois que la connexion est lancée
Various arguments are also found in the position of direct object for the verb “activer”:
Activer: un canal, la mémoire, le débogage, la transmission, le cache, une imprimante.
Support verbs or full verbs are also collected in this way:
Faire un lien, créer un lien, établir un lien/ maintenir le lien, faire pointer un lien, supprimer un lien, modifier un lien
In other words, our methodological approach allows translators to create a term base that is complete and completely corpus-driven.
Potential terms, among others, including the unit resolution in the subject area of digital cameras are: low resolution, high resolution, standard resolution. Should these three candidates be listed as collocations or multiword units? In the context of specialized translation, an important criterion consists in examining the way the French translations of these collocations behave:
low resolution faible résolution
high resolution haute résolution (“weak” resolution)
standard resolution résolution standard
The position of the adjective in French, as well as other different syntactic behaviour points to the conclusion that the first two can be listed as terms, whereas the third must be considered as a collocation. This situation thus led me to try and answer some questions about the definition of collocation within a French Linguistics paradigm. In some theoretical approaches adopted within French Linguistics (theoretical approaches), the limits between a collocation, a “frozen” expression, or a support verb and its noun predicate are quite unclear.
The gap between listening to someone explaining what wonderful results corpora can lead to, and actually plunging into a corpus can be quite huge. Most students in linguistics or translation studies are discouraged when faced with a corpus and a concordancer. The systematic approach described here helps to guide people through an ocean of data, and to develop a fine-tuned linguistic intuition.
Uta Lenk (Universität Augsburg, Germany)
Fixed or stabilized expressions, phrases, collocations or idioms are the subject of an ever-increasing number of linguistic studies. The question of the (graded) degree of stability of such expressions, and their importance within the vocabulary of a language is now undisputed. However, satisfying explanations for the nature of the gradation displayed in their fixedness and/or variability remain to be found.
The current project, based on the BNC and various spoken corpora of English, is an investigation of the phraseological patterns of the high-frequency lexeme time and searches for such an explanation. A multitude of patterns that include the word time, displaying differing degrees of fixedness, have been identified and will be demonstrated.
The fact that these patterns display different degrees of fixedness justifies a broad categorization into three main groups that could be called ‘frozen expressions’ at one end of the scale of fixedness, ‘semi-fixed expressions’ in the middle and ‘stabilized expressions’ at the other extreme.
While ‘frozen expressions’ are lexically invariable (such as time and time again, by the time, from time to time), ‘semi-fixed expressions’ allow for a limited range of lexical variation in a certain syntactic pattern (i.e. a long/short time, at a/any/one/some (modified) time or (the) N of time). ‘Stabilized expressions’, finally, are verbal patterns that allow for certain syntactic variation as they may encompass a range of optional and/or obligatory slots that may be filled with various valency requirements such as pronouns, or they may even include other frozen and semi-fixed expressions (such as spend time, take time, have time).
Between these three groups, but also within, different tendencies regarding their collocational behavior are noticeable. The stabilized expression spend time, for example, displays markedly different patterns of distribution regarding not only the different types of the verb, but also the different possible combinations with prepositional attachments (i.e. spent time in, spends time with). Semantic aspects also contribute to these patterns of distribution.
The stabilized expression take time comes with several meanings, each of which display their own syntactic requirements, such as an 'empty subject requirement' in it usually takes a long time to ..., whereas have time is more frequently than not.used with a negation
The differences in pattern appearances of the various ‘stabilized expressions’ that have been identified are associated with the variability of syntactic aspects on the one hand, but must also be seen in connection with an as yet undescribed semantic variability of the node, time, on the other hand. For linguists, dictionary makers and language teachers, the question arises as to how detailed an analysis of collocational tendencies is indeed desirable and/or feasible, especially in the light of an approachable and learnable definition for foreign learners.
Gunter Lorenz (Universität Augsburg, Germany)
It can hardly be disputed that, over the last fifteen years or so, the availability of machine-readable language corpora has fundamentally changed the discipline of descriptive linguistics. And while corpus linguistics has sparked off great progress in almost all areas of the description of English, it has also served further to deconstruct one of the most persistent myths of language description and teaching – that of a homogenous ‘Good English’ standard.
The pre-sociolinguistic concept of ‘Good English’, as agreed on by the linguistic authorities and laid down in the standard grammars and dictionaries, was an idealisation based on careful, educated, formal English usage. Such usage is of course included in present-day English language corpora, but they are in no way restricted to it. The 100 million word British National Corpus (BNC), for example, consists of a wide range of spoken and written genres, with data from speakers and writers of all ages and from a wide variety of social and regional backgrounds. The more we study such rich variation, and the more we learn about the wealth of variants, the more arbitrary a monolithic standard of English will appear. In a corpus-based description of English, the formerly cast-iron certainties of standard grammar often need to be replaced by probabilities – or transformed into meticulous descriptions of linguistic and extra-linguistic conditions and restrictions. This way even ‘macro-rules’ of grammar are reduced in scope, and even the most frequently cited grammatical rules of English can be found to be ‘violated’ in actual, native-speaker usage.
This state of affairs, however, is not altogether new: learners have for a long time tantalised teachers with counter-evidence for grammatical rules from pop songs, the media or cryptic dictums from a native-speaking friend. English language corpora, while posing a far greater challenge to grammatical description than such accidental data, have also provided a partial remedy: they have allowed us to research, rather than criticise, the seeming violations of ‘rules’; they have enabled us to isolate and investigate, rather than marvel at, the variability inherent in English grammar and usage – and they are forcing us to re-evaluate the concept of a monolithic standard of English.
As a sequel to a paper given at TaLC 2000, the present contribution will continue the thread of probing into the variability of perfective and progressive aspectual marking. As far as the perfective is concerned, there will be a brief discussion of the relationship between use of the perfect and adverbial perfective marking – through items such as yet, hitherto and since, for example. This relationship, it will be argued, is not nearly as stringent as some grammars of English (especially pedagogical) have held so far.
As regards the progressive, there have recently been a number of studies concerned with its variability and its increase in frequency (see, e.g., Mair & Hundt 1995, as well as Smitterberg, Reich & Hahn 2000). Explanations have so far been of a rather general nature, emphasising what has been termed the ongoing ‘colloquialisation’ of English, and maybe even language use in general. While this interpretation certainly cannot be rejected from a wide perspective, it does not explain precisely what governs the change from individual simple to progressive variants. The present paper seeks to offer an additional view: by narrowing the focus on selected members of two related classes of verbs, namely those of ‘inert perception’ and ‘inert cognition’ (cf. Leech 1987: 24f), it will try to show how new variants are construed in their individual instantiations. It will be argued that the progressive offers new construals for these verbs and the meanings they express – in one case (see) leading to a new Aktionsart, and in another (think) leading to a re-grammaticalisation of a lexicalised structure.
Returning to the problem of ‘deconstructing’ standard grammar rules, corpus evidence such as the present indeed works towards questioning the validity of variation-free macro-grammar. But it would be wrong to hold corpus linguistics responsible for the deconstruction of the standard: language corpora have not deconstructed the grammar, but the concept of grammatical rules without variation. And there is a strong case for the assumption that such rules do not in fact exist. Now the Pandora’s box of linguistic description has been opened, the best that can be achieved is to explore ever new types of variants with the aim of understanding the nature of linguistic variation. For the present purpose, for instance, conventional corpus data has been supplemented with what may be perceived as examples of a particularly ‘sassy’, i.e. inventive, type of usage from the media. It is in such ‘trendy’ usage that change and variation become most apparent – not always of a permanent nature, but certainly of interest to linguistic description. In relation to the main theme of the present ICAME conference, one of the ‘future challenges for corpus linguistics’ might include, on a practical level, incorporating more – and more recent – dynamic audiovisual material into corpus research. On a more theoretical level future challenges include reconciliation of corpus linguistics with other disciplines, such as cognitive linguistics and grammaticalisation theory.
Leech, G. (1987) Meaning and the English Verb. Longman, London.
Mair, Ch. and Hundt, M. (1995) Why is the progressive becoming more frequent in English? A corpus-based investigation of language change in progress, Zeitschrift für Anglistik und Amerikanistik 43(2): 111-22.
Smitterberg, E., Reich, S. and Hahn, A. (2000) The present progressive in political and academic language in the 19th and 20th centuries: a corpus-based investigation, ICAME Journal (24): 99-118.
Fanny Meunier (Université catholique de Louvain, Belgium) & Inge de Mönnink (University of Nijmegen, The Netherlands)
The aim of our research project is to assess the performance of an automatic part-of-speech tagger (namely the TOSCA-ICLE Tagger Lemmatizer1) on learners’ written productions in English, taking test data from the International Corpus of Learner English (ICLE2). Once the performance of the tagger has been precisely assessed and analysed, we will first define and formalize the influence of the learner’s mother tongue on the performance of a tagger originally designed for and trained on native English. Secondly we will customize the tagger by implementing a number of changes (either probabilistic or rule-based) in order to improve its success rate on learner material.
To answer the question of whether or not it is valid to annotate learner data with a tagger trained on error-free native English material, we have selected nine 5,000 word sub-corpora from ICLE, representing 9 different mother tongue backgrounds: French, Dutch, German, Polish, Spanish, Italian, Russian, Swedish and Finnish. These sub-corpora were tagged automatically and the output was manually checked.
In assessing the performance of the tagger, we classified the errors into three broad categories:
The performance of the tagger was lower than the 95% claimed in earlier research (e.g. de Haan 2000). This difference may be due to the fact that for the current assessment the tags were compared in detail, including punctuation, and taking every single subclass into account.
More striking perhaps is that, contrary to our initial expectation, the majority of the errors were not due to the non-nativeness of the input, which constitutes an encouraging factor for learner corpus tagging. Thanks to the probabilistic component of the tagger, erroneous input was often tagged correctly.
A high percentage of the tagger errors we found can be resolved by adding a few simple rules to the tagger. The output of the new tagger should then be assessed anew and it is our firm belief that the tagger will score significantly better, not only on learner data but also on native data.
Aarts, J., Barkema, H. and Oostdijk, N. (1997) The TOSCA-ICLE tagset. Tagging manual. Nijmegen, TOSCA Research Group.
Granger, S. (1996) Learner English around the world. In Greenbaum, S. (ed.) Comparing English Worldwide. Clarendon Press, Oxford, pp.13-24.
Granger, S. (1998) The computer learner corpus: a versatile new source of data for SLA research. In Granger, S. (ed.) Learner English on Computer. Addison Wesley Longman, London and New York, pp.3-18.
de Haan, P. (2000) Tagging non-native English with the TOSCA-ICLE tagger. In Mair, C. and Hundt, M. (eds) Corpus Linguistics and Linguistic. Proceedings of the 20th ICAME Conference, Freiburg 1999. Rodopi, Amsterdam, pp.69-79.
Dieter Mindt (Freie Universität Berlin, Germany)
For thousands of years grammarians have stated the regularities of languages in the form of grammatical rules. We have become used to the concept of ‘grammatical rule’ without asking questions such as
· · What is the internal structure of a grammatical rule?
· · When can we be confident of having discovered a grammatical rule?
· · What is the place of exceptions and errors in the description of language?
· · What inferences can be drawn for language change from the structure of a grammatical rule?
Corpus linguistics sheds some new light on these questions. A fundamental research design consists basically of three steps: collecting data, classifying data, drawing conclusions from the classified data in the form of grammatical rules. The procedure is inductive (from language to grammatical generalisation), rather than deductive (from pre-stated rule to example).
The process requires among other things the definition of grammatical categories together with a descriptive framework outlining the essential features of the grammatical phenomenon under investigation taking into account morphologic, syntactic, and semantic information. These features can be described in the form of variables which take a number of different values.
The final step is to draw conclusions from the classified data in the form of grammatical rules. If all previous procedures have been carried out properly we arrive at the standard form of a grammatical rule.
There is one important feature which characterises the standard form of a grammatical rule. Of the many possible realisations, only between two and four make up the core of an individual grammatical phenomenon. The paper describes the form of the standard result using examples from morphology, syntax and semantics. A clear borderline can be drawn between grammatical rules and the behaviour of lexical elements.
The data provide a new perspective on cases that are traditionally described as exceptions, errors, obsolete uses and emerging tendencies.
The paper concludes with a discussion of the following objectives
(1) (1) to achieve a closer approximation to the concept of "grammatical rule"
(2) (2) to give new insights into the processes of language change.
Joybrato Mukherjee (University of Bonn, Germany)
At the ICAME conference in 1998, Jürgen Esser outlined a new approach to the linguistic description and analysis of prosody-syntax interactions in spoken English. His suggestions largely capitalise on Halford's (1996: 33) concept of a talk unit as the "maximal unit defined by syntax and intonation". For various reasons, her definition, however, remains somewhat vague. This leads Esser (1998: 481) to redefine the talk unit more precisely as a ‘stretch of speech which, at a given point, is syntactically complete and ends with a falling tone’. The present paper reports on some of the general results of a research project in which the modified talk unit model was applied in the annotation of a 50,000 word sample corpus in order ‘to analyse empirically how intonation and syntax come into operation along with each other in authentic language use’ (Mukherjee 2001: 151).
The corpus material was mainly taken from the London-Lund Corpus of Spoken English (LLC) and complemented with texts from a small corpus of monologues including texts read aloud. All tone unit boundaries were annotated by indicating the prosodic status (non-final in case of a rise; final in case of a fall) and the syntactic status (non-final in case of syntactic incompleteness; final in case of syntactic completeness). With regard to the syntactic status, some finer distinctions were made, depending, for example, on whether a syntactic structure is completed later in the text or broken off in mid-sentence. In general, the talk unit ends whenever both the prosodic and syntactic channel show a final status at a given tone unit boundary followed by a new syntactic beginning to the right. Due to the fact that the talk unit is defined both prosodically and syntactically it is regarded as a parasyntactic presentation structure: at the level of parasyntax, syntax and intonation are integrated (cf. Mulder 1989: 90), and by means of stylistic choices at the level of parasyntax the speaker presents his or her message to the hearer. The combination of prosodic and syntactic status at a given tone unit boundary is called a parasyntactic configuration. Differences in their use across the corpus is referred to as parasyntactic variation.
The quantitative corpus analysis unveils significant and, at times, surprising correlations between the parasyntactic variation and the stylistic (including text-typological) variation across the corpus. The wealth of the statistical evidence vindicates the general assumption that the talk unit represents a stylistically relevant unit of presentation. Thus, the overall conclusion may be drawn that talk units fulfil a communicative function.
First and foremost, talk units function as information structural units. A talk unit comprises one to many tone unit(s). According to Halliday (1994: 295), the tone unit ‘is not only a phonological constituent; it also functions as [...] a unit of information in the discourse’. In applying this information structural interpretation of tone units, the functional corpus analysis reveals that speakers tend to pack their message into contour-defined chunks of information so that a tone unit indeed corresponds to an information unit. On the basis of this parasyntactic information packaging, speakers choose parasyntactic configurations at tone unit boundaries in such a way that an efficient and effective hierarchy of tone units (as information units) is established. Generally speaking, the parasyntactic configuration at a tone unit boundary signals to the hearer the relative weight or importance of the subsequent tone unit. So, the tone unit boundary serves as a kind of window through which the hearer may look forward onto the next information unit (cf. Esser 1993: 144). The in-depth analysis of authentic corpus data shows that both talk units as such and their internal structure (in terms of segmentation and hierarchy of tone units) makes it easier for the hearer to process the information presented by the speaker.
Secondly, the corpus analysis shows that not only information structural functions can be ascribed to talk units, but that talk units turn out to be relevant to speaker interaction in conversations, too. In particular, talk unit boundaries provide appropriate positions for turn taking. In this context, the grade of politeness can be analysed and categorised on the basis of (prosodic and/or syntactic) completeness at the very point of speaker shift. Hence, the concept of talk units also contributes to a better understanding of some important linguistic principles of turn taking which have been vaguely described in previous approaches. This also applies to pausological research which has often concentrated on the description of the demarcating function of pauses at tone unit boundaries. Conversely, the talk unit-based approach highlights the fact that many pauses in authentic discourse are used as ‘information structural means in order to increase the hearer’s sense of anticipation whenever syntactic incompleteness and/or prosodic openness is given’ (Mukherjee 2000 in press).
Esser, J. (1993): English Linguistic Stylistics. Niemeyer, Tübingen.
Esser, J. (1998): Syntactic and prosodic closure in on-line speech production, Anglia 116: 476-91.
Halford, B. (1996): Talk Units: The Structure of Spoken Canadian English. Narr, Tübingen.
Halliday, M.A.K. (1994): An Introduction to Functional Grammar. Second edition. Arnold, London.
Mukherjee, J. (2000): Speech is silver, but silence is golden: some remarks on the function(s) of pauses, Anglia 118, in press.
Mukherjee, J. (2001): Form and Function of Parasyntactic Presentation Structures: A Corpus-based Study of Talk Units in Spoken English. Rodopi, Amsterdam, in press.
Mulder, J. (1989): Foundations of Axiomatic Functionalism. De Gruyter, Berlin.
JoAnne Neff, Emma Dafouz, Honesto Herrera, Francisco Martínez, Juan Pedro Rica (Universidad Complutense de Madrid, Spain), Mercedes Díez (Universidad de Alcala, Spain), Rosa Prieto (E.O.I. Madrid, Spain) & Carmen Sancho (Universidad Politecnica Madrid, Spain)
This paper, part of the work for a project funded by the Spanish Ministry of Education (BFF2000-0699-C02-01), presents the results of a contrastive study of qualification devices used in a 400,000-word corpus of English argumentative text, produced by EFL Spanish university writers (SUW, from the International Corpus of Learner English, ICLE, Louvain), American university writers (AUW, from the LOCNESS Corpus, Louvain), and native professional writers (NPW, newspaper editorials in English, the English-Spanish Contrastive Corpus, Madrid).
We use the term qualification to indicate the means the writer uses to express to the reader whether the information presented is to be understood as fact or opinion (Hyland 2000). By devices, we mean the grammatical and lexical means used to construct writer stance, defined as ‘…the positioning of a social agent with respect to alignment, power, knowledge, belief, evidence, affect and other socially salient categories’ (Du Bois 2000). Stance involves, then, not only the writer’s expression of commitment, but also the effect that the writer presupposes the stance taken up will have upon the readers, thus necessitating the use of politeness strategies.
In a previous study (Neff et al. in press), we examined certain modals of probability and reporting verbs, as used by Spanish EFL writers and American university writers to construct writer stance (Biber & Finnegan 1989). The findings regarding modal verbs showed significant differences in the uses of can, may, and might, but not of could. For the reporting verbs, results showed that SUW overused say and underused state. Along with the overuse of we can, the under- and overuse of some reporting verbs seemed to be working against impersonalization strategies in the Spanish EFL texts.
In the present study, we attempt to explore these previous findings by carrying out a further analysis of the same modals and reporting verbs in the argumentative texts of the SUW (194,845 words), the AUW (149,790 words) and an additional third corpus of professional writers of newspaper editorials, the PNW (113,475 words). In addition to the previously mentioned modals, we decided to study non-epistemic must (e.g., we must consider) for possible use as a positive politeness strategy (cf. we can). Results from these data are contrasted with those from other ICLE sub-corpora (Dutch, French, German, and Italian) for interlanguage issues and, for developmental questions, with those from the MAD Corpus (first- and fourth-year Spanish EFL university students writing in both L1 and L2).
Both this study and the previous one have suggested that Spanish university writers construct a different stance from that created by native writers. Below, we summarize our conclusions.
1) Some of the problems that SUW experience may be due to typological differences between Spanish and English, that is L1 factors, for instance, an overuse of can influenced by the use of poder in Spanish. However, there may be, as well, developmental factors to consider, for example, in the SUW’s overuse of we can and we must as stance- or discourse-markers, but not of we may and we might. Since can is the first modal verb learned in the Spanish EFL classroom, NNS may feel comfortable using it – in the assumption that it covers the same degrees of doubt as poder (can) in Spanish – and, thus, do not risk using other English modals, such as may and might.
2) The differences which appeared in the previous study in relation to the sociocultural conventions used in constructing writer stance were confirmed in this study. The SUW overuse of we can and we must followed by verbs of mental and verbal processes points to a transfer of politeness strategies from the Spanish academic context. The SUWs’ use of reporting verbs also further contributes to the less impersonal writer’s role. The comparison of reporting verbs for the three groups showed similar total tokens, but notably different frequencies for individual verbs. This suggests that SUW not only concentrate on a limited set of verbs, which restricts their possibilities of modulating their statements, but also do not use inanimate subjects, which would allow for more abstract agency. Such strategies point to broader issues of the interactional patterns based on positive-politeness used in peninsular Spanish, that is -power/-distance, while English may use, globally, more negative politeness strategies, that is –power/+distance.
3) We are well aware that some of these differences, particularly between the PNW corpus and those of the student writers, may be a result of genre characteristics, given that editorials are a very controlled type of text. Nevertheless, we believe that non-native texts should not be compared solely to native student texts, which may display some developmental characteristics not present in more sophisticated writing.
Arja Nurmi (University of Helsinki, Finland)
In this paper I look at the item must in Early Modern English Correspondence. The corpus used is the 2.7 million word Corpus of Early English Correspondence (CEEC). The time period covered will be the whole span of the CEEC (1410?–1681).
An earlier study (Nurmi forthcoming), using the 0.45 million word Corpus of Early English Correspondence Sampler (CEECS), showed that the frequency of the item must does not greatly change during the 16th and 17th centuries, staying mainly in the 10–20 instances /10,000 words range. This is slightly lower than the results for Drama in the ARCHER corpus in the17th century, but is clearly higher than the frequency for News or Fiction (Biber et al 1998: 208). Earlier studies (e.g. Nurmi forthcoming) have shown that the results from the full CEEC as opposed to the CEECS are likely to lessen the fluctuation between subperiods, but on the whole follow the trends attested in the sampler.
The reliability of the CEECS was shown once again, when a total of 2858 instances of must were retrieved from the full CEEC: the range of variation and the general trends in the development of must remained more or less within the picture suggested by CEECS.
When the full CEEC was divided into three subcorpora (15th, 16th and 17th centuries) a slight increasing tendency in the frequency of the auxiliary could be observed, from 7.2 instances of must /10,000 words in the 15th century to first 8.1 in the 16th and 13.2 in the 17th century.
Men and women both show the rise in the frequency of the item must, but women’s use of the auxiliary increases faster than men’s (women start from 4.2, move to 7.9 and finally to 15.6, while men start from 7.8, move to 8.1 and in the 17th century to 12.2). The difference in frequency between men and women is not very great, however.
The main focus of this paper is on the sociolinguistic and socio-pragmatic variation associated with a construction not part of an ongoing change. Most previous work in the field of historical sociolinguistics has focused on changes in progress. This time, it is interesting to see the variation connected with an item only very slightly increasing in frequency for nearly three centuries. One of the aims is to study the different meanings of must and their association with social variables. This is to be a pilot study into the feasibility of such an undertaking, and, if successful, will function as the model for a later application of the same method to other modals as well.
The model of meaning applied to must is the fairly simplified one found in Biber et al (1999). The instances of must are classified under only two meanings:personal obligation and logical necessity. In a corpus-based study, it is necessary to have only few meaning categories, otherwise, when connected to sociolinguistic variables, the results would suffer from what Rissanen (1989: 18) calls “The Mystery of Vanishing Reliability”. The application of a model based on the Present-day English meanings of must is not without its problems, but preliminary results seem to indicate that this model can be applied to Early Modern English data without doing great injustice to the material.
Examples of logical necessity meaning of must in the CEEC:
- - he that hath nobody to work for him, must keepe shopp himself (John Holles, 1625)
- - you must thincke we are brought to a lowe ebbe when the last weeke the archdukes ambassador was caried to see the auncient goodly plate of the house of Burgundie (John Chamberlain, 1613)
Examples of personal obligation meaning of must in the CEEC:
- - All the hedges and fences must be allso presently made. (John Holles, 1630)
- - I must confess this sodaine allteration of your purpose and promise makes me imploye my patience and dewtie; (Thomas Barrington 1629).
Since women’s use of the item must seems to increase more rapidly than that of men’s, it will be particularly interesting to see what kind of meanings are used by men and women. One factor that may have some bearing on the matter is the relationship between letter writer and recipient. In the case of personal obligation it is also interesting to see who the person obliged to do something is: the writer, the recipient or a third person.
Biber, D., Conrad, S. & Reppen, R. (1998). Corpus Linguistics. Investigating Language Structure and Use. Cambridge University Press, Cambridge and New York.
Biber, D., Johansson, S., Leech, G., Conrad, S. and Finegan, E. (1999). Longman Grammar of Spoken and Written English. Longman, London.
Nurmi, A. (forthcoming). Modal auxiliaries in EmodE correspondence. Paper presented at the 11th International Conference on English Historical Linguistics. 7–11 September 2000, Santiago de Compostela, Spain.
Rissanen, M. (1989) Three Problems Connected with the Use of Diachronic corpora, ICAME Journal 13: 16–19.
Nelleke Oostdijk (University of Nijmegen, The Netherlands)
Spoken language corpora present interesting new challenges to researchers who are working on the construction of parsers. While over the past years efforts have been directed predominantly at the implementation of parsers for the linguistic annotation of written corpora, the idiosyncracies of spoken language data have been largely neglected. The provisions that were made in order to deal with the spoken passages in written fiction should be considered for what they are, viz. ad hoc provisions that were made to prevent the parsing process from being upset and forced to a halt. Meanwhile, in the light of the absence of parsers geared to the analysis of spoken language, it is not surprising to find that parsers that were originally constructed for parsing written language data are being applied for the analysis of spoken data. The overall performance of the parser on such data will prove very poor. What we see then is that in order to accommodate the parser, language data are being normalized, i.e. they are made to conform to what are postulated to be the ‘rules of grammar’. Such opportunistic doctoring of the data is opposed to the view generally upheld in corpus linguistics that the data are autonomous.
The present paper seeks to investigate a number of phenomena that are considered to be characteristic of spoken language use, in particular disfluencies such as hesitations, false-starts and self-corrections. The aim is to get insight in the nature, frequency and distribution of these phenomena, so that we may consider the implications this has for the construction of a parser geared towards the analysis of spoken language data. The study is based on the normalized data found in the spoken part of the parsed ICE-GB corpus.
ICE-GB, the British component of the International Corpus of English (Greenbaum 1996) comprises some one million words of spoken and written English produced by adult, educated, native speakers of British English. The texts in the corpus date from 1990-1994 inclusive, i.e. all texts were originally published or recorded during this period (Nelson 1996a). The corpus has been fully tagged for part-of-speech information, while it has also been syntactically annotated. In the annotation process the TOSCA-ICE parser was used (Oostdijk 2000).
Prior to the linguistic annotation of the corpus, all the material – both spoken and written – was marked up, using two types of markup: (1) textual markup, which was added to the texts themselves and typically encodes features of the original text that are lost when it is converted into a computerized text file, and (2) bibliographical and biographical markup, which was stored externally in the form of a file header for each text (Nelson 1996b: 36). While it was observed that “spoken texts, and especially dialogues require much more markup than written ones”, the set of (textual) markup symbols was designed also to include symbols that could be used “to indicate such features as pauses, speaker turns and overlapping segments” (ibid. 39-42). Nelson’s discussion of the various markup symbols, however, reveals that quite a number of them do not so much preserve features that would otherwise be lost, but were introduced on the grounds that without these markup symbols automatic parsing would be problematic:
This system for marking overlaps was adopted because complete speaker turns are essential for parsing. The marking scheme indicates the overlapping without making the turns discontinuous.
(Nelson 1996: 41)
Especially with the spoken data from the corpus, textual markup has been applied to normalize the input for the parser:
Spoken English is characterized by a wide range of nonfluencies which are not found in writing. (…) These phenomena are transcribed as they occur, and the markup for them will be of particular interest to researchers studying the interaction between speakers. However, they may be seen as disruptions of the underlying syntax, and as such are problematic from the point of automatic parsing. We use the general term 'normalization' to describe the method of using markup to deal with them.
Normalization involving normative deletion is applied, for example, with repetitions, self-corrections, and hesitations ‘when they disrupt the syntax’. Items marked for normative deletion do not form part of the input to the parser.
Normalization involving normative insertion includes the use of markup to complete the input. Here the role of the annotator/linguist is rather objectionable since completing what was left incomplete by the speaker can only amount to speculation. The same goes for the normative insertions which aim to ‘correct’ the utterance. Here the annotator/linguist imposes his own subjective norms as to what is syntactically, semantically, or otherwise correct English by deleting what he finds erroneous and then inserting what is considered to be correct.
While Nelson points out that the “principle has been to normalize the original text as little as possible, and to do so only when it was essential for parsing”, I want to argue here that instead of using textual markup to modify the data whenever one sees fit, a solution for the problems encountered in trying to parse spoken language data must be found in obtaining insight in the nature of the phenomena and the problems they pose for automatic parsing. Here the normalized data from the spoken part of the ICE-GB corpus can serve as a starting-point.
From the 600,000-word spoken subcorpus slightly over 10,000 instances were collected where the original input had been normalized. In the data 15 categories were represented, ranging from direct, spontaneous conversations to scripted speeches. First an inventory was made of the average number of instances per text category. The results do not immediately suggest a relation between the text category on the one hand and the frequency of normalization on the other hand. Here, however, the idea comes to mind that this might well be an artifact of the way in which the textual markup has been applied. Next, on the basis of what helpful information could be found in the literature (eg Fromkin 1973a,b, 1988; Garrett 1988, Nooteboom 1973; Tanenhaus 1988), a classification scheme was developed that should serve to distinguish between different types of disfluencies. The scheme was then applied to the data. The picture that emerged is one which suggests that in many instances adaptation of the parser is feasible so that there is no need to normalize the data prior to parsing.
Fromkin, V. (1973a). The non-anomalous nature of anomalous utterances. In V. Fromkin (ed.), pp.144-163.
Fromkin, V. (1973b). Appendix. In Fromkin, V. (ed.), pp. 243-69.
Fromkin, V. (ed.) (1973). Speech errors as Linguistic Evidence. Mouton, The Hague.
Fromkin, V. (1988). Grammatical aspects of speech errors. In Newmeyer, F.J. (ed.) (1988a), pp. 117-38.
Garrett, M. (1988). Processes in language production. In Newmeyer, F.J. (ed.) (1988b), pp. 69-96.
Greenbaum, S. (1996.) Introducing ICE. In Greenbaum, S. (ed.), pp. 3-12.
Greenbaum, S. (ed.) (1996) Comparing English Worldwide. The International Corpus of English. Clarendon Press, Oxford.
Nelson, G. (1996a). The design of the corpus. In Greenbaum, S. (ed.), pp. 27-35.
Nelson, G. (1996b). Markup systems. In Greenbaum, S. (ed.), pp. 36-53.
Newmeyer, F.J. (ed.) (1988a). Linguistics: The Cambridge Survey. II Linguistic Theory: Extensions and Implications. Cambridge University Press., Cambridge.
Newmeyer, F.J. (ed.) (1988b). Linguistics: The Cambridge Survey. III Language: Psychological and Behavioral Aspects. Cambridge University Press, Cambridge.
Nooteboom, S. (1973). The tongue slips into patterns. In Fromkin, V. (ed.), pp. 144-63.
Oostdijk, N. (2000). Corpus-based English linguistics at a cross-roads. English Studies. A Journal of English Language and Literature 81(2 ): 127-41.
Tanenhaus, M. (1988). Psycholinguistics: An overview. In Newmeyer, F.J. (ed.) (1988b), pp. 1-37.
Pascual Pérez-Paredes (Universidad de Murcia, Spain)
In computer programming, integration is combining software or hardware components or both into an overall system; in everyday language it is the act or process of making whole or entire. In this work, we want to do exactly that, in short, to present a framework for corpus data gathering and implementation that integrates state-of-the-art technology and network multipoint approaches into a learning environment. To this purpose, a description of tools and procedures will be presented.
As Computer Assisted Language Learning (CALL) spreads, educational institutions and students are becoming more familiar with the use of computers in foreign language learning. These days, CALL environments are progressively expanding their capabilities and functionalities and, thus, it is not uncommon to find Universities or Secondary Schools in Europe and the States where different learning formats (Jackson 2001) combine to provide teachers and students with rich and varied learning experiences.
Simultaneously, learner language corpora have become an important resource for both linguists and teachers, not surprisingly, for a wide variety of reasons. It is claimed that learner corpora can provide language teaching professionals and language researchers with a thorough insight into the language actually used by foreign language students. FLT professionals and linguists are very much interested in using corpora, and thriving symposia such as TALC are a token of this interest. However, exclusively oral corpora of students’ foreign language are scarce. Written language is, no doubt, still dominant in the field. Very recently, Basturkmen (2001) has claimed that ELT has focused its attention on describing and teaching the written language and talks about spoken language as being neglected.
In this piece of research, we set out to explore ways of incorporating oral learner corpora into mainstream CALL environments within a technology-enhanced e-learning approach, as this has been the basis for our research and our primary source of learner feedback. This environment is characterised by (1) the pressence of the teacher in the computer facility; (2) the fact that the sessions are, to different degrees, live, face-to-face and instructor-led and (3) the use of materials which have somehow been previously delivered to students as they are familiar with the type of tasks underlying the corpus. Typically, this environment is asynchronous. A technology-delivered e-learning approach was not considered, as our learner audience and the instructor himself actually met to share the learning sessions.
At least six domains of interest have determined our drive towards compiling and using a corpus of spoken learner English or any other foreign language. The following reasons are common to both written and spoken corpora:
1. 1. A corpus can contribute to a better understanding of students’ use of the foreign language.
2. 2. It can offer teachers classroom data that are not frequently analysed, fundamentally because classroom dynamics make it extremely difficult for teachers to monitor every learner’s performance. Usually, teachers tend to concentrate on fragmented chunks of discourse and, more often than not, students are too aware of this monitoring task, which in different ways can make learners shy away from otherwise natural use of the FL. Such a corpus gives teachers the chance to examine performances of students in detail, both as individuals and as a group.
In addition, learner oral corpora (LOC) might prove useful in at least the following ways:
3. 3. Learners’ oral performance can be diagnosed and measured based on both qualitative and quantitative information. This way, teachers can form a second opinion on their student’s output and both on-the-spot and continuous assessment can be conveniently enhanced.
4. 4. Group assessment is favoured. Teachers can establish accurate performance comparisons between two individuals, between two groups of students or even between learner and native speakers’ corpora. This is extremely useful when it comes down to deepening into students’ oral output, a territory often neglected by educators and learners themselves presumably believing that the very nature of oral discourse is unapprehendable and elusive.
5. 5. Teachers can build monitor learner corpora that can contribute to changes and adjustments in their methodology, particularly those aspects that are more directly connected with developing students’ oral skills.
6. 6. LOC can be used to promote students’ language awareness of both segmental and suprasegmental aspects of learners’ FL production.
All six purposes might potentially determine different compilation, annotation, if applicable, and access criteria. In this sense, Granger (1998: 8) lists some of the features that are relevant to learner corpus building, distinguishing between language and learner variables. In the first group we find medium, genre, topic, technicality and task setting; in the second we find age, sex, L1, region, other foreign languages studied or spoken, L2 level, learning context and practical experience. These features are to be carefully considered and weighed up when designing learner corpora. It seems that compilers have two options here. The first is to opt for heterogeneous samples of learner language and account for them on different levels, mainly those of representation, tagging and codification (Llisterri 1999: 54); the second is to strive for homogeneous collections of texts in terms of the two broad categories presented above. This continuum is to be examined by corpus builders in the light of the purposes and expectations generated by the corpus.
Our research adheres to the second approach as we believe functional parameters to be absolutely necessary when design stages of an oral learner corpus are first sketched out. As any corpus serves a purpose which must be conveniently kept to (Sánchez et al. 1995), our proposal here is to link learner oral corpora to their expected audience, functionality, access technology and network delivery system. In the main, we will discuss three major types of learning formats that can be conveniently adapted to learner oral corpora implementation: asynchronous self-study, synchronous instructor-led events and, third, group collaboration. These formats will combine with the purposes outlined above offering a wide range of potential activities for the FL classroom. The principle underlying our discussion is that Computer Based (CBT) and Instructor Led Training (ILT) can both play a decisive role in boosting the use of customized oral learner corpora in language teaching institutions. In the same manner, this work aims at increasing the relatively small amount of research into computer networking and language learning (Kern and Warshauer 2000).
Susan Pintzuk & Ann Taylor (University of York, United Kingdom)
The fact that English changed from a predominantly OV to a strict VO word order language over time is well known. Less is known about exactly how this change came about. Van der Wurff (1999) makes the interesting observation that the change did not affect all NPs at the same time; rather, positive NPs (that is, non-negative, non-quantified NPs) lose the ability to appear in preverbal position before quantified and negative NPs. Van der Wurff's interpretation of this difference relies on a grammatical reanalysis at the beginning of the 15th century. Using a Kaynian head-initial framework, he assumes that the derivation involves movement to specAgrO; but the result is the same if an underived OV base order is adopted instead, as he himself notes (van der Wurff 1999: 253, fn.8). At the point of the reanalysis, the original derivation for OV word order is lost, but an alternative derivation becomes available for quantified and negative NPs: namely, quantified NPs move leftward over the verb by overt quantifier raising, while negatives move leftward to specNegP. Thus, under van der Wurff's hypothesis, all NPs behave alike up to the 15th century; afterwards, negative and quantified NPs behave differently from positive NPs.
There is, however, an alternative analysis available for this data, namely that negative NPs, quantified NPs and positive NPs behave differently right from the beginning of the Old English period. Kroch and Taylor (2000) and Pintzuk (2000) have suggested that in both Old and Middle English there is grammatical competition in the underlying structure of the VP, VO vs. OV, with the frequency of VO structure gradually increasing at the expense of OV structure, until OV is completely lost in Late Middle English. The synchronic evidence from Old and Middle English demonstrates that there is a distinct difference in the derivation of positive preverbal objects as opposed to negative and quantified preverbal objects: that is, preverbal positive NPs have only one derivation, base-generation, while preverbal negative and quantified NPs may either be base-generated or scrambled leftward from postverbal position. Consequently, van der Wurff's claim that there was no difference in the behaviour of the three types of NPs prior to the reanalysis cannot be correct.
In this paper we use data from three corpora, the Brooklyn Corpus and the York-Helsinki Corpus for Old English and the Penn-Helskinki Corpus for Middle English to examine in more detail the long-term effect of NP type on the course of the change from OV to VO. During the Old English period, there is a decrease in the frequency of preverbal positive objects which is strongly influenced by the length of the NP, suggesting that most objects are base-generated preverbally with postposition of heavier objects. Negative NPs, on the other hand, show little if any decrease in preverbal frequency, and quantified NPs fall somewhere between the two. In Early Middle English, there is a similar decrease in frequency of preverbal NPs, but only negative and quantified NPs show a length effect; positive NPs, in contrast, show no length effect. This suggests that, unlike in Old English, the vast majority of positive NPs are base-generated postverbally. Moreover, the frequencies of the three types decrease at different rates. This can be interpreted as the converse of the Constant Rate Effect (Kroch 1989) and provides strong evidence that the derivations of the three types are different at least from the beginning of the Middle English period. After all, if the three types are derived in the same way, we would expect the frequency of preverbal position to decrease at the same rate for all three.
In conclusion, we have shown that the quantitative diachronic evidence supports the earlier qualitative synchronic evidence that positive, quantified and negative NPs do not all behave in the same prior to the 15th century, as van der Wurff's analysis predicts.
Antoinette Renouf (University of Liverpool, United Kingdom)
A previously unimaginable number and range of electronic text corpora are now available to corpus linguists, ranging from small and sampled collections to very large textual databases. Whilst this wealth of data makes possible many types of corpus-based research, particularly in the formerly rather inaccessible areas of lexis and lexico-grammar, it has inherent limitations. One is that the appropriate corpus data and software may not be available at the required moment without special arrangements having been made - access to the right computer, appropriate licences having been obtained, and so on. Other limitations are the size, modernity and static nature of the corpora, which can preclude certain kinds of linguistic empirical investigation, for instance the study of very rare, new or changing language features.
An alternative source of linguistic information is the web, a publicly available data resource containing a vast and evolving accumulation of texts. Admittedly, this is not contructed or managed with the rigour or for the purpose of a corpus. It is a muddle of multilinguality; it operates a loose definition of ‘text’ which includes all manner of extraneous matter; text dating is sporadic and linguistically uninterpretable, so that neither the latest coinages nor the elements of language change across time that are undeniably in there are identifiable by means of chronological organisation. Nevertheless, as a renewable resource which in itself costs the linguistic community nothing to create or access, it is worthy of serious consideration.
The web offers immediate and free online access to some large English corpora. Both the BNC and Bank of English provide free response to word and phrase searches. As Rundell (2000) says, the output is ‘deliberately limited in order to encourage you to opt for the full package’, and even then, ‘you will only get back 40-50 corpus lines for any inquiry’. Nevertheless, as he concludes: ‘these sites are a good starting point for the occasional corpus query. Even 50 contextualized examples of going to would be a good basis for some useful hypothesis-testing’.
The web is both larger than any corpus and constantly updated. One way to exploit its potential as a linguistic resource is to extract texts which, together, make up a corpus deemed to be representative of something, such as ‘general language use’, or a technical field. Kilgariff (2001) et al., for instance, are currently collecting reference sets of urls, from which domain-specified corpora which the user can create by downloading, without copyright infringement.
Another approach is to exploit the functionality of web search engines. Standard engines operate by searching the web for information containing a specified search term. A small effort of imagination recasts this in corpus linguistic terms as searching the web for contexts containing a target word or phrase. A growing number of researchers have been driven, by the absence or insufficiency of evidence in existing corpora for rarer or newer linguistic items and features, to attempt a trawl of the web by this means. Search engines are, however, not designed to accommodate such an approach, and the consequent negotiation entails tedious serial searching and downloading of individually thin pickings, followed by painstaking manual editing of whole texts. Each search engine covers only a slice of the web, and only retrieves search terms in its periodically-updated index. Google, which is unique in extracting context for search terms, cannot trace Sophiegate, of April 2001 vintage, by early May. As a linguistic search tool, its results are also skewed, in including only one instance of a search term from a given web page on which it may actually occur several times.
A further leap of imagination reveals the web to be ripe for exploitation by software tailored to find and retrieve contextualised instances of words and phrases. Such information could serve the linguistic community in areas of linguistic, pedagogic, lexicographic and other endeavour by filling the information gaps left by traditional corpus data. This was the point of departure for the WebCorp project, launched in December 2000 in the Unit at Liverpool. The project team consists of Mike Pacey and Andrew Kehoe, software developers; Paul Davies, statistical consultant; and myself, linguist and project originator and manager. Michael Hoey, as PL, is on hand with linguistic advice; Themis Bowcock, as CL, steers us through web and post-web developments. The WebCorp tool is being developed according to an intensive and ambitious two-year project plan, which has been informed in part by the copious feedback received in response to a simple prototype software demonstrator installed on the web in May 2000. Among the many unsolicited expressions of enthusiasm for the WebCorp tool, Michael Rundell stated in his paper entitled ‘The biggest corpus of all’ (Rundell 2000) that:
...a major breakthrough is at hand, in the form of a stunning new website that produces real "concordances". As with Altavista and others, http://www.webcorp.org.uk/ [i.e. WebCorp] searches the entire Internet for your query. But in this case the output is a proper concordance with an amount of surrounding context which the user (that's you) can specify in advance. The results, in other words, look very similar to what you might get from the BNC or COBUILD Direct - but in this case the "source data" is the vast store of text on the entire Internet.
Basically, the WebCorp tool interfaces with the user request, converting it into a format acceptable to a selection of existing search engines. It then piggy-backs on one or other of these that has been specified by the user. Each search engine follows its own procedure for searching a section of the web for texts containing the specified language item. Once the engine has traced the search term, via its own index, to a candidate text, WebCorp downloads that text into memory and extracts the appropriate linguistic context, processing and collating it before presenting it to the user. This basic functionality is relatively simple to achieve. The real challenges lie in developing a closer understanding of the web's structure and content, and in devising ways of compensating for the current limitations of search engines in order to produce a maximally efficient, informative and user-friendly tool.
Progress is incremental but swift. Since the prototype tool was first reported on (Renouf 2001), a new version has emerged, with improvements including smaller font and compact presentation for concordance lines, numbered concordance lines, and HTML-centred keywords. The next version (4.7) of the tool will soon incorporate type/token counts for web pages, contiguous collocational statistics, and improvements in speed of search and retrieval. WebCorp presently allows the user to specify search engines/s, data source and concordance format; by way of illustration I offer below a 10-word context, HTML-formatted concordance extract retrieved via the Northern Light search engine at midday on May 2nd, 2001.
Sample WebCorp output for the recent neologism, Sophiegate:
1. isn't likely to end "Sophiegate" soon. Word is, the newspaper
2. called R-JH. Thanks to Sophiegate , she's stepped down and
3. the sharp end of the Sophiegate skewer. Tony Blair put on
4. Britain's closet republicans. Sophiegate is a huge blow to
5. monarchy faces tough choices over "Sophiegate" tapes. Apr 06 2001 17
6. interest between the two. The "Sophiegate" affair will also be a
7. to say that the recent "Sophiegate" scandal involving Sophie Rhys-Jones
8. lucrative deal. The so-called "Sophiegate" scandal led many newspaper editors
9. Pak-origin scribe set up `Sophiegate'. Pope commemorates Good Friday
10. isn't likely to end "Sophiegate" soon. Word is, the newspaper
11. column Marketing & PR|Press & publishing 'Sophiegate': what the papers say
12. 2001. Mark Lawson on the Sophiegate. 31 Mar 2001. Mark Lawson
In this particular case, WebCorp is able to extract up-to-the-month results for a vogue formation; it can equally yield instances of usage which is stable but rare, too rare to appear in established corpora. The extract serves to demonstrate that the web, when accessed by WebCorp, offers linguistic evidence that is not supplied by existing text corpora.
We acknowledge the support of this research by EPSRC, with thanks.
Kilgarriff, A. (2001). Web as corpus. In Rayson, P, Wilson, A., McEnery, T., Hardie, A. and Khoja, S. (eds) Proceedings of the Corpus Linguistics 2001 Conference; UCREL, 2001, pp.342-344.
Renouf, A. (2001). The Web as a source of linguistic information. In Rayson, P., Wilson, A., McEnery, T., Hardie, A. and Khoja, S. (eds) Proceedings of the Corpus Linguistics 2001 Conference ; UCREL, 2001, pp. 492
Rundell, M. (2000). The biggest corpus of all. Humanising Language Teaching 3 (May 2000) (http://www.hltmag.co.uk/may00/idea.htm).
Geoffrey Sampson (University of Sussex, United Kingdom)
Most British children reach school speaking English fluently. If all goes well they complete compulsory schooling as skilled users of the written language. The compilation of structurally annotated corpora is starting to open up new possibilities of studying the trajectory taken in moving from one stage to the other.
Our recent CHRISTINE and current LUCY projects have been compiling annotated corpora of, among other genres:
* conversational speech
* published writing
* writing by 9-12 year old children
All samples are annotated according to exactly the same scheme, defined in Sampson (1995). This paper discusses some initial findings extracted through statistical comparison of these samples.
Writing "wordier" than speech
A first expectation is that words in published writing are likely to be organized into more elaborate constructions, on average, than words in spontaneous speech. We cannot examine this in terms of "average sentence length", because sentences are not well-defined units in speech; instead, I examined mean length in words of all constructions:
child writing 7.68
published writing 9.45
The average construction in published writing is about twice as long as the average spoken construction, and the average construction in child writing is intermediate.
Width v. depth in parse-trees
Ignoring details of individual constructions for a moment and thinking just about the abstract geometry of parse-trees, there are two ways in which mean construction length can differ. Branching can be wide, and it can be deep. By wide branching I mean that a tagma may have many ICs (daughter nodes); for instance, "a man" is a noun phrase with two daughters, "a funny little man [who made us laugh]" is a phrase with five daughters. By deep branching I mean the extent to which structures exploit the recursive nature of grammar to create long chains of branches between words and root nodes. A tree with deep branching will dominate many words even if each nonterminal node is only two ICs "wide".
Width and depth are not mutually exclusive, and one might expect both to be relevant to genre differences. That turns out to be wrong. Average ICs per construction are:
child writing 2.860
published writing 2.759
-- essentially no differences.
For depth, things are different. To make my figures for depth as theory-neutral as possible, I have counted depth of words in terms of the number of clause nodes between the word and the parse-tree root. Mean word depths are:
child writing 1.401
published writing 1.857
These figures perhaps do not look very different, but that is a consequence of the way depth is counted. You cannot have a subordinate clause without words in a matrix clause introducing it, so increasing the overall depth of embedding in a parse-tree does not increase the average word depth in proportion. But these differences are very solid. On a one-tailed t-test, the difference between word depths even as between speech and child writing (where the mean difference is smallest) gave a significance statistic far beyond the critical value for p < 0.0001, the largest critical value I could find.
That does not mean that the large difference shown above between mean depth in speech and in published writing is wholly attributable to the difference between linguistic modes. We know from earlier work (Sampson 2001) that mean depth of grammatical structures increases with age, not just in childhood but throughout adult life. The average published writer is probably older than the average in the whole population, so the authors of the published writing may be individuals whose speech structures were relatively complex. But the published writing v. speech differential shown above is about seven times greater than the difference, within the speech sample, between middle-aged speakers and the overall average.
Thus it seems that the difference in "wordiness" between published writing and speech relates to greater use of recursion, and in this respect the children's writing has moved part-way from the spoken towards the adult written norms. Let us now examine specific constructions.
I used the chi-squared test to test for significant differences between incidence of various categories of tagma within the total set of tagmas in the three samples. In the following, an entry such as "sp < cw = pw" means that, for the given grammatical category, the rate in speech is significantly lower than in child writing, but there is no significant difference between the child-writing and published-writing rates. ("Significant" means p < 0.05, but in fact each significant difference achieves at least p < 0.01.) "x >> y" means that the rate for x is over 50% greater than for y.
noun phrase sp << cw = pw
verb group sp >> cw < pw
prep. Phrase sp << cw < pw
adjective phrase sp < cw = pw
adverb phrase sp = cw = pw
number phrase sp > cw >> pw
determiner phrase sp >> cw = pw
genitive phrase sp << cw = pw
In each case but number phrases, the child writing figure is closer, often much closer, to the published writing than to the speech figure. In the area of phrase grammar it seems that the children have moved quite a long way along the path of adaptation to the norms of ‘model’ written prose.
For subordinate clause categories the story changes. (Here, for some of the less-frequent categories, significant differences attain only the p < 0.05 but not the p < 0.01 level.)
The commonest subordinate-clause type is the infinitival clause: its frequency is very similar in the three samples and what differences occur are non-significant.
Otherwise (ignoring two types which are very infrequent in all genres) subordinate clauses can be classified in four groups.
(1) nominal clause sp >> cw << pw
verbless clause sp >> cw < pw
antecedentless relative sp >> cw = pw
bare non-finite clause sp >> cw = pw
These clauses are less frequent in published writing than speech, and the child-writing figure is closer to published writing: one might say that children have successfully learned to ration their use.
In the case of nominal and verbless clauses, this rationing has "overshot" in the sense that the child-writing usage is even lower than in published writing. I cannot explain the figures for nominal clauses; for verbless clauses I suspect that the children have taken thoroughly to heart teachers' injunctions to "write in complete sentences".
(2) adverbial clauses sp = cw >> pw
Child writing reflects the relative high usage in speech.
(3) present participle clauses sp = cw << pw
comparative clauses sp = cw = pw
"with" clauses sp = cw = pw
special "as" clauses sp = cw << pw
Child writing reflects the relatively low usage in speech, or there are no significant differences between child writing and either of the other genres.
Groups (1)-(3) are "easy" cases: either children's writing resembles spontaneous speech with respect to the figure in question, or the children have learned to do something less often than in speech. The "harder" case, presumably, is where published writing shows a higher incidence than speech, and the children's writing is closer to the former:
(4) relative clause sp << cw = pw
whiz-deleted relative sp << cw = pw
past participle clause sp << cw < pw
These are, logically speaking, variants of one construction. Apparently, children at the relevant age have learned to adapt to adult written norms (where adaptation means using more than in conversation rather than fewer) earlier in the case of relative clauses (and their logical equivalents) than with other subordinate-clause types.
Anyone can understand why adaptation might be faster for phrases than clauses: phrases are simpler. But, among clauses, one might feel that relatives are more complicated than other types, not less complicated. So these findings are not what one might expect.
The above analysis is obviously extremely broad-brush and crude. Most of what we would like to know about children's acquisition of literacy relates to far more detailed properties of language. But those features too, or many of them, are available for study in this range of annotated corpora. When one first begins to extract analytic findings from a body of data, one naturally examines the most obvious and immediate properties. The above is not a very subtle analysis. But it is a beginning.
Sampson, G.R. (1995) English for the Computer. Clarendon Press, Oxford.
Sampson, G.R. (2001) Demographic correlates of complexity in British speech. Ch. 5 of Sampson, Empirical Linguistics. Continuum.
Josef Schmied (Chemnitz University of Technology, Germany)
This presentation uses a contrastive translation corpus and a (contrastive, deductive) grammar (book) on the web as a basis for contrastive linguistic and language learning research. It discusses the sociobiographic questionnaire, the tracking mechanisms and the first results of the experiments on students’ behaviour in the Chemnitz Internet Grammar. It uses prepositional phrases as an example and compares the deductive explorations and the inductive searches for corpus samples and the work on the exercise component. Although prepositions in German and English appear rather similar on a general quantitative level, some specific analyses show that particularly non-prototypical and metaphorical usages can be quite different. This analysis offers some concrete examples of linguistic problems and resulting learner strategies, distinguishing between several user types, like age, computer literacy and other variables that might have an influence on the choice of learning style. In this context, I will also compare the emphasis of linguistic and psychological perspectives of ‘explorative learning as work’ and present further plans on the Chemnitz Internet Grammar.
Kristina Schneider (University of Rostock, Germany)
Having assembled a corpus of English newspapers from 1700 to 2000 (the Rostock Newspaper Corpus or RNC), we have now started to extend the corpus to cover German newspapers (and later perhaps newspapers published in other European languages) to permit contrastive studies.
The initial design and analysis of the German corpus will be based on the selection principles underlying the English corpus, which were presented at ICAME 1999 in Freiburg (cf. Schneider 2000a), but various modifications are envisaged for later stages of the corpus compilation.
Overall corpus design
Period: 1700 to 2000 in 30-year intervals
Sample size: 10,000 word samples from 6 newspapers per period
Corpus-lines: Distinction between popular and quality papers
For the initial stage the German corpus will cover the same period, consist of the same sample intervals, the same sample sizes and will attempt the same differentiation between popular and quality papers.
Selection of newspapers for the corpus
The selection will be based on 'newspaper profiles' using the following criteria for the distinction of popular and quality papers: external criteria (circulation and price) and internal criteria (news content, i.e. percentage of hard news and soft news; non-news content such as advertisements and serialisations; and layout, e.g. decorations, titles and sub-titles).
Judging from a pilot study of German 18th and 19th century newspapers, there seems to be a certain division between popular and quality papers, although this distinction is probably less pronounced than in the English press at the time. Most early German newspapers still concentrate on hard news (politics and business) and occasionally show instances of borderline cases, but seem to contain few clear-cut instances of soft news (human-interest material such as natural disasters, accidents and crime). This contrasts with English newspapers of the same period which quite frequently include soft news. Another factor which might make it difficult to spot popular papers is that the early German samples show fewer decorations and less spectacular titles and sub-titles than the respective English newspapers.
The corpus analysis of the English corpus proved that differences between popular and quality papers are also reflected in the language used (sentence length, paragraph length, word length, choice of vocabulary, including forms of address and names), and that it is the popular papers which usually introduce important new trends: trends towards better readability (e.g. short and less complex sentences) and more emotional involvement (e.g. frequent use of buzz words and superlatives). These parameters will also be applied to the German corpus, but may have to be changed to account for the morphological bent of the German language compared to English.
As in the investigation of the English corpus, special attention will be paid to the development of headlines. In particular, we will examine the claims put forward by Wölfle (1943: 411) and Sandig (1971: 139, 142f), namely that German newspapers – compared to their English counterparts - lagged behind in the use of headlines, and that the first headlines were devoted to political events (e.g. 1848 Revolution in Paris, 1870 Franco-German War) and not to soft news items as in our English samples. It will also be interesting to compare the use of nominal and verbal headlines (or regional and relational headlines, cf. Schneider 2000b) in German and English newspapers, since the German language tends to prefer nominal constructions (cf. Sandig 1971: 147, 157).
Differing political and socio-economic conditions
Contrastive studies thrive on differences. As far as English and German newspapers and their readership are concerned, these differences are abundant as proved by the following selection:
1. 1. Whereas the English press was to a large extent concentrated in London, the German newspaper landscape faithfully reflects the political fragmentation in pre-Bismarck times and later the federal structure of German society. Therefore, it is impossible to expect a concentration of newspapers in one and the same centre for all three centuries under consideration, as is the case with London. Since regional papers have always been much more dominant in Germany than in Britain, our focus will have to be on different centres of newspaper publishing in Germany, such as Leipzig, Cologne, and later Berlin. Even today, 93% of the German daily press are composed of regional newspapers (Schulz 1997: 47).
2. 2. Although for the initial stage of corpus collection the starting point of 1700 has been taken over from the British press, which experienced a relatively constant and free development after the abolition of the Licensing Act in 1695, German newspapers have been confronted with several ups and downs in their development since 1700. They have experienced frequent changes of periods of strict censorship (absolutist regimes, Napoleonic military administration, Nazi regime, Communist regime in East Germany) and periods of more liberal attitudes (Revolution of 1848, the Press Act of 1874 ['Freiheitliches Reichspressegesetz'], and finally press freedom in post-war West Germany and, more recently, East Germany).
3. 3. Another factor which may have delayed the development of the distinction between popular and quality newspapers in Germany is that industrialisation started about 30 years later and, consequently, a large working class readership came into existence later than in England.
The corpus will not only provide a basis for contrastive work, it will also fill a gap between the two corpora of German newspapers that we are aware of: the Tübingen and the Bonn projects. The former concentrated on 17th century German weeklies such as the Relation (founded in Straßburg in 1605) and Aviso (founded in Wolfenbüttel in 1609) (cf. Fritz/Straßner 1996, Schröder 1995), the latter was concerned with a comparison of East and West German newspapers of the second half of the 20th century (cf. Hellmann 1992).
Fritz, G. and Straßner, E. (eds) (1996) Die Sprache der ersten deutschen Wochenzeitungen im 17. Jahrhundert. Niemeyer, Tübingen.
Hellmann, M. W. (1992) Wörter und Wortgebrauch in Ost und West. Ein rechnergestütztes Korpus-Wörterbuch zu Zeitungstexten aus den beiden deutschen Staaten. Die Welt und Neues Deutschland 1949-1974. Narr, Tübingen.
Sandig, B. (1971) Syntaktische Typologie der Schlagzeile. Möglichkeiten und Grenzen der Sprachökonomie im Zeitungsdeutsch. Hueber, München.
Schneider, K. (2000a) Popular and quality papers in the Rostock Historical Newspaper Corpus. In: Mair, Ch. and Hundt, M. (eds), Corpus Linguistics and Linguistic Theory. Papers from the Twentieth International Conference on English Language Research on Computerized Corpora (ICAME 20). Rodopi, Amsterdam, pp. 321–37.
Schneider, K. (2000b), The emergence and development of headlines in British newspapers. In: Ungerer, F. (ed.) English Media Texts: Past and Present. Benjamins, Amsterdam, pp. 45-65.
Schröder, Th. (1995), Die ersten Zeitungen. Textgestaltung und Nachrichtenauswahl, Narr,. Tübingen.
Schulz, V. (1997), Medienkundliches Handbuch. Die Zeitung, 5. aktualisierte und überarb. Neuauflage. Hahner Verlagsgesellschaft, Aachen-Hahn.
Wölfle, L. (1943), Beiträge zu einer Geschichte der deutschen Zeitungstypographie von 1609-1938. Versuch einer Entwicklungsgeschichte des Umbruchs, Dissertation, München.
Noëlle Serpollet (Lancaster University, United Kingdom)
Corpus linguistics is a methodology which now uses bilingual corpora as practical tools both to test theoretical hypotheses in contrastive analysis and to throw light on particular translation problems. Hence the linguistic approach is a very important aspect of translation studies. Bilingual parallel corpora, also called translation corpora, are defined in Baker (1995: 230) as being composed of “original source language texts in language A and their translated version in language B”. Corpus linguistics also uses more and more parallel synchronic or diachronic corpora (such as LOB/Brown or LOB/FLOB) to study the possible influence of one type of English on another (American English [henceforth AmE] on British English [BrE]) or the evolution of linguistic features.
This paper will focus here on an area, where until recently, little contrastive analysis had been carried out. I will analyse the particular grammatical features that are mandative constructions, in two genres ‑ Press and Learned Prose ‑ of French and English. Example (1) below illustrates a French mandative subjunctive and its translation into English by a mandative construction with should:
La formidable menace que présente la prolifération nucléaire et balistique dans le monde exige, en effet, que toutes les précautions soient prises. (Le Monde, 1993)
:: The daunting threat of nuclear and missile proliferation in fact requires that every possible precaution (should) be taken. (The Guardian Weekly 1993)
Example (2) provides a mandative construction with should and a subjunctive as its French equivalent:
However, it is preferable that these high-speed channels should, as far as possible, be placed […]. (International Telecommunication Union)
:: Toutefois, il est préférable que les voies à grande rapidité de modulation soient dans la mesure du possible, établies […]. (ITU)
The analysis of selected texts extracted from the bilingual parallel INTERSECT corpus (Salkie 1995) will be undertaken using ParaConc, a bilingual parallel text concordance program (Barlow 1995). The French section was previously tagged with Cordial 6 Universités. I will first work from French into English (Le Monde [1992-93] translated in The Guardian Weekly) and study the English equivalents of the French subjunctive in the Press category. Then I will go back to French and analyse the Learned Prose category of INTERSECT, tracking the mandative constructions in English and their translations in French.
I will check the validity of my findings (from the English section of INTERSECT) and see how significant they are, by comparing them with the results obtained in the thorough analysis of extracts, equivalent in size, date and categories, of the one-million-word corpora LOB and FLOB. The Press category is composed of originals (FLOB) and target texts in BrE (INTERSECT). I will deal here with comparable corpora, defined as follows in Baker’s sense (1995: 234): two separate collections of texts in the same language A (BrE), one corpus containing original texts in that language and the second containing translations from a source language B (French) into the language A. The corpora are the same length and cover the same genre(s). Hence, the possible differences in the results could be due to the translation process. The Learned Prose category contains originals (FLOB) and source texts in BrE (INTERSECT). Here, any difference would be due to the data themselves as no translation process is involved.
This analysis will enable me to describe the evolution of mandative constructions from the 1960s to the 1990s in two genres and to see if two corpora of modern BrE show the same trend regarding this specific grammatical feature.
The results will show (like previous ‘corpus-based investigation of language change in progress’ (Mair & Hundt 1995): Övergaard 1995 and Hundt 1998) that the mandative subjunctive is increasing whereas mandative should is on the decrease. However, these studies used to some extent, non-computerized, incomplete or incomparable corpora.
The originality of my research lies in the fact that it involves developing complex queries in Xkwic (Christ 1994) to retrieve only the relevant occurrences of both the modal and the subjunctive. Therefore, all my results are totally comparable because I used exactly the same retrieving queries on complete, grammatically tagged and computerized corpora.
Nonetheless, these findings do not answer the following question: If the mandative subjunctive is indeed healthy in British English, what explanation can be provided? One possible answer (suggested in many observations on American vs. British English, see also Serpollet ) would be that its health is sustained by the influence of American English.
With four carefully matched corpora now available (thanks to the completion of the one-million-word corpora FLOB [Freiburg-LOB, 1991 – that has been used along LOB in the second part of this paper] and Frown [Freiburg-Brown, 1992], the two 1990s counterparts of the 1960s LOB and Brown), an exhaustive corpus-based study of language change in progress over a thirty-year period can be conducted. This study can analyse and compare synchronic corpora to examine for example the possible influence of American English on British English with FLOB/Frown.
The final part of this paper will attempt to verify the explanation mentioned above by analysing two corpora of American English: Brown and its 1992 counterpart Frown. The analysis of American data will enable me to see if the ongoing change in BrE is dependent of diachronic developments in AmE or of the synchronic influence of AmE.
Baker, M. (1995) Corpora in translation studies: an overview and some suggestions for future research, Target 7(2): 223-243.
Barlow, M. (1995) ParaConc: a Concordancer for parallel texts, Computer and Text 10: 14-16, Oxford University Press, Oxford.
Christ, O. (1994) A modular and flexible architecture for an integrated corpus query system, COMPLEX'94, Budapest.
Hundt, M. (1998) It is important that this study (should) be based on the analysis of parallel corpora: On the use of mandative subjunctive in four major varieties of English. In Lindquist, H. et al. (eds) The major Varieties of English, Papers from MAVEN 97, Växjö University.
Mair, C. and Hundt, M. (1995) Why is the progressive becoming more frequent in English? – A corpus based investigation of language change in progress, Zeitschrift für Anglistik und Amerikanistik 43: 123-132.
Övergaard, G. (1995) The Mandative Subjunctive in American and British English in the 20th Century, Stockholm, Almqvist & Wiksell International, Acta Universitatis Upsaliensis, Studia Anglistica Upsaliensia, Vol. 94.
Salkie, R. (1995) INTERSECT: a parallel corpus project at Brighton University, Computer and Texts 9: 4-5, OPU.
Serpollet, N. (2001) The mandative subjunctive in British English seems to be alive and kicking… Is this due to the influence of American English? In Rayson, P., Wilson, A., McEnery, T., Hardie, A. and Khoja, S. (eds) Proceedings of the Corpus Linguistics 2001 Conference, (Lancaster University, 30 March-2 April 2001), Lancaster, UCREL, (Unit for Computer Research on the English Language : technical papers, Volume13 - Special issue).
Robert Sigley (Daito Bunka University, Japan) & Janet Holmes (Victoria University of Wellington, New Zealand)
This paper uses data from the LOB, Brown, FLOB, Frown and WWC corpora to investigate the uses (and especially, potentially sexist uses) of the terms girl(s) and boy(s). To the extent that language reflects and even constructs social reality, studying representative samples of language use (i.e., corpora) provides a useful means of tracking social change in progress. The corpora used in this analysis allow us to follow changes in several varieties of English between 1961 and 1991.
An initial survey shows that the overall frequencies of girl(s) and boy(s) are lower in Frown and FLOB than in the earlier corpora. This apparent decrease has several sources: an orthographic shift from boy(-)friend to boyfriend; a possible shift towards use of non-gender-marking terms (the overall frequency of child(ren)/ kid(s) shows an apparent increase); and a possible shift away from use of both terms for adult referents.
A more detailed breakdown of use in context shows that there has been a shift of the last type, albeit a quantitatively small one. There is a continued asymmetry in use of girl for adults (more than three times more likely than parallel use of boy in all corpora), but the use of boy with adult reference has fallen from 21% in LOB to 17% in FLOB (10% in WWC), while use of girl for adults has fallen from 61% in LOB to 51% in FLOB (38% in WWC). Use of both terms in workplace contexts shows a similar decrease.
Inspection of concordance data indicates that both girl and woman are used to refer to young but sexually mature human females, with the choice between these determined less by objective age than by a cluster of subjective connotations including immaturity, innocence, youthful appearance, desirability, subordinate status, domestic roles, and emotional dependence or vulnerability.
Analysis of the collocates of the terms girl(s) and boy(s) in the corpora largely confirms this picture, but also offers some evidence of incipient social change. The distribution of age premodifiers for girl(s) favours young girl, which in practice most frequently denotes an adolescent or adult, whereas boy(s) are more closely associated with small, used for younger children. There is also particular emphasis on describing girls’ appearance; but this imbalance is less marked in Frown and FLOB. Finally, verbs of which girl is an object show a shift from meet, marry, love, find, know (LOB), married (Brown) to teased (FLOB), which appears consistent with a movement from older to younger referents, and away from girl as an object of male desire.
Overall, our results suggest that, unlike adult males, adult females continue to be linguistically constructed as immature, with special emphasis on related properties traditionally attractive to men, including a youthful appearance, a subordinate or submissive role, and emotional dependence. The most optimistic interpretation of our data is that there is some slight evidence of the beginnings of a decrease in the patterns analysed. The WWC writings emerge as more ‘socially progressive’ than FLOB, a pattern consistent with all other sexist language variables so far compared.
Our paper concludes with an evaluation of the strengths and weaknesses of corpus analysis in this area of language and gender research. Corpora of writing provide a valuable window on the usage trends received by language consumers, and may reflect ongoing social trends; but the reflection of social change provided by numerical analysis of published writings is a distorted one, for several reasons.
Firstly, these corpora are not ideally designed for sociological research. In particular, writer gender was not controlled in constructing any of these corpora. The apparently more feminist tendencies of the WWC data probably derive in part from it having a larger proportion of female writers than the other corpora (to the extent this can be estimated: e.g. it is the only one of these corpora to have a female majority in its fiction texts) — though this in itself could indicate social progress has been made.
Secondly, corpus figures provide little direct attitudinal information. Sexist patterns may be used ironically, or put into the mouths of characters by authors who may not themselves support those usages. Many of the examples cited in this paper are of this type. Since individual tokens may in fact be subverting an apparently sexist usage, it is thus critically important to consider how items are used in context.
Thirdly, in these relatively small 1-million-word corpora a single text can have a hugely disproportionate influence on raw frequencies and collocate lists. This could be minimised by using larger corpora; but editorial influence and prescription can lead to an even broader clumping of variant choices in any corpus of writing. As a result, corpus research is necessary but not sufficient; complementary research into actual spoken usage, and attitudes, and prescription, is also needed.
Nicholas Smith & Geoffrey Leech (Lancaster University, United Kingdom).
The focus of the presentation will be work carried out so far on recent change in grammatical usage in the English verb phrase, especially aspect and modality. The primary data for the investigation are the one-million word LOB and FLOB corpora of British English, sampled in 1961 and 1991-2 respectively.1 These data have been recently complemented by two smaller spoken corpora, extracts from the Survey of English Usage and ICE-GB corpora, sampled across roughly the same time period.2
As was demonstrated in several papers at ICAME 2000, across the regional varieties of English there have been a number of interesting shifts in overall frequency of some of the modal auxiliaries. For example, must, shall, need and ought to, and to a lesser extent may, might and would have undergone significant declines. The general pattern seems to be one of a contraction in the profile of modal verbs as a whole. This is only partly compensated by slight increases in some of the so-called semi-modals, notably have to and need to. Moreover, in the corresponding spoken corpora of British English, ICE-GB-mini and SEU-mini, although the type distributions differ somewhat, the overall directions of change are the same.
To explore this further, we have investigated the semantic, syntactic and contextual features of those modals whose distribution has changed the most. Adopting the semantic categories developed in Coates 1983, it has emerged that proportional and total frequencies of the deontic/root senses of must and may have drastically declined, whereas with should it is the epistemic use that has declined, while the deontic/root use has remained robust. Minority usages in 1961, such as the quasi-subjunctive uses of may and should, have also declined markedly. With regard to the semi-auxiliary constructions, prima facie one of the most salient characteristics is the rise of have to. This may be attributable to a number of factors, such as have to taking over semantic territory from must (which, for the obligational reading, may be felt to be more direct, authoritative, assertive etc.), the rise of epistemic have to, or fulfilling syntactic functions not available to must (e.g., use in past tense, or following an infinitive marker). On the other hand the spread of have to does not occur across all genres; its rise across LOB/FLOB is pronounced in the press genres only, whereas the fall of must occurs across all 15 text categories. Furthermore, the rise of have to is much more pronounced within direct speech quotations than without.
Much as been said about the long-term historical rise of be going to to express the future (see e.g. Mossé 1938, Visser 1973, Krug 2000). Across LOB and FLOB, however, a continuation of this growth has not been detected. Meanwhile, the SEU and ICE-GB data show this to be one area in which speech represents a significant point of divergence.
In continuation of work presented at ICAME 2000, we will also present further analysis and discussion of changes in the English progressive construction, focussing on present progressives and modal combinations with the progressive, as these have been found to be the main areas of increase in the thirty year period.
News headings before 1800 often appear to be static and clumsy in comparison to the informative summary headline of the 20th century. They are frequently referred to as functional labels that neither advertise nor summarise the content of the news. Typical examples of functional labels are shown in (1) and (2):
(1) (1) Daily Advertiser Hague Intelligence 1741CEA00179
(2) (2) Foreign Affairs. 1741CJL00758
But there are also more elaborate headlines, such as (3) and (4):
(3) Now on Sale in any Quantity at Ashley Lee & Co.: Brandy Warehouses, on Ludgate-hill, A Parcel of fine ORANGE SHRUB, 1741LDP02076
(4) On Monday, the 2nd of February, will begin A Course of Anatomy. 1741CEA00179
Of the existing studies in news language most are concerned with the emergence of the modern headline of the 20th century. The present paper focuses on earlier types of headings found in the first newspapers (between 1670 and 1800) and is part of a larger project which aims to classify news items in the newspaper collection of the Zurich English Newspaper Corpus.
The ZEN Corpus – a selection of the most widely read London newspapers of the late 17th and of the 18th centuries – consists of newspaper issues manually entered into a database. One of the central concerns while keying in was to group individual news items together according to a set of preliminary criteria. The criteria relevant to this first classification were the news categories established by the editors of the newspapers themselves. The following text types have been identified so far (in alphabetical order): Accidents, Address, Advertisement (commercial, personal), Announcement, Births, Crime, Deaths, Essay, Foreign news, Home news, Letter, Lost and found, Proclamation, Review, Ship news, Weddings.
Although helpful for a first sampling of related text types, this procedure also shows major flaws and inconsistencies. Often there is no clear-cut dividing line between two news items, and the classification, which mingles functional, formal, and hierarchical levels of analysis, leads to overlapping news categories. One way to remedy the situation now is to revise the existing classification through systematically categorising the newspaper headings. News headings are immediately accessible dividing lines of textual units; they signal a juncture and indicate the content of the subsequent news.
The goal of this paper, therefore, is twofold: Firstly we wish to present plausible ways of approaching a systematic classification of news headings in the ZEN Corpus. Secondly, we will tackle technical problems encountered in this process of classification and present solutions for the whole corpus.
For the former purpose a selection of newspapers from the ZEN collection was isolated and analysed. The years used for this primary investigation – 1701 and 1741 – include a comparable range of material. A minimum of two sample years was selected so as to possibly cover the full variety of different headings found in the corpus. The 40 year gap between the two samples is to make sure the results show a diachronic development.
Following the sampling a number of features were assigned to each heading. A fundamental distinction had to be made between headings introducing editorial news and those heads preceding advertisements. On the basis of this distinction each heading was further specified by a cover term indicating the content of the news (e.g. product advertisement, etc). In a next step additional criteria were included which cover general (graphical, positional), structural (morphological, syntactic), and contextual (situational and socio-historical) factors.
Whereas a general and formal classification can easily be achieved through internal analysis of the headings, the study of contextual factors requires knowledge of the tradition of the genre. A few traditional text types have been identified and described so far (cf. Sandig 1972; more recently Schneider 2000). To these categories more types have been added which – we believe – can clearly be distinguished. The analysis of headings finally leads to establishing concepts of ‘norm’, ‘deviation’, ‘style’, and ultimately to the description of ‘house styles’ of individual newspapers included in the ZEN collection.
In the technical part of our presentation, we will demonstrate the use of a flat-file database program for annotating headline features. The headlines as marked up in the original text version of the corpus were transferred to a Filemaker database containing the preliminary set of descriptive features.
One of the advantages of Filemaker is that it makes it possible to change and add fields as well as descriptions without changing the rest of the database contents. All text internal headline features were entered as binary factors, i.e. each criterion can be thought of as an answer to a specific yes-no question. This system allows a finely grained characterisation of headline types. Identifying co-occurring factors (and exact opposites) will enable the collation of factors into labels without overlapping features. In addition to the text internal "binary" factors, a number of interpretative features will be added. The final feature set will then be incorporated into the corpus in XML form.
Sandig, B. (1971) Syntaktische Typologie der Schlagzeile. Möglichkeiten und Grenzen der Sprachökonomie im Zeitungsdeutsch. Max Hueber Verlag, München (Linguistische Reihe 6).
Schneider, K. (2000) The Emergence and Development of Headlines in British Newspapers. In Ungerer, F. (ed.) English Media Texts Past and Present. Language and Textual Structure. John Benjamins Publishing, Amsterdam., (Pragmatics and Beyond 80), pp. 45-65.
Irma Taavitsainen, Päivi Pahta & Martti Mäkinen (University of Helsinki, Finland)
We report on progress in compiling the Corpus of Early English Medical Writing (CEEM) and present some of the latest results obtained in pilot studies using the corpus. Work on the corpus was started in the Department of English at the University of Helsinki by Irma Taavitsainen and Päivi Pahta a few years ago, with the aim of compiling a computerised database for the project Scientific Thought-styles: The Evolution of Early English Medical Writing. We first presented the corpus plan in ICAME 1996, and have given papers based on corpus findings in ICAME conferences in 1997-1999. The project is now funded by the Academy of Finland (1999-2001), and the team has been joined by Martti Mäkinen, working on his dissertation1, and undergraduate students working on their Master’s theses.
The present size of the corpus is c. 1.5 million words. The medieval part extends from 1375 to 1550 and is further divided into two subperiods (1375-1475 and 1475-1550). It contains c. 560,000 words and is nearing completion. The text selection covers the full range of medical texts from the first emergence of writings in this register in English. The material contains vernacular translations of academic tracts, surgical and anatomical treatises, texts in special fields like ophthalmology, encyclopaedias and compendia, remedybooks and recipes, and medical verse. Shorter texts are included in toto, and more comprehensive treatises are represented by extracts. This part opens up new possibilities for studying vernacularisation, the creation of a new prestige register in English, and the processes of standardisation, to name a few central topics of interest for linguistic studies.
Work in charting scientific and medical writing in the early modern period has already been carried out. The text selection for the latter part of the corpus 1550-1660 is nearly complete; but the last subperiod (1660-1750) still needs complementing. We aim to cover the widening use of English in medical writing in this period. The material ranges from specialized texts and experimental reports published in the Philosophical Transactions of the Royal Society of London to popular health guides intended for the general audience. The widening spectrum provides material for comparison; we have included statutory texts dealing with hygiene, religious and moral treatises on diseases, and educational works giving guidelines for healthy living. The conventions of scientific writing were already established in English in this period, and it is possible to discern various lines of development in the genres within this register of writing.
The corpus has already been used for pilot studies by the project members, and we hope to make a version of it publicly available with the other historical corpora compiled in Helsinki. Copyright negotiations have been initiated for this purpose. We have prepared a WordCruncher version, and the Corpus Presenter by Prof. Raymond Hickey is now being tested. For more information and publications related to the corpus and the project, see our homepage at:
Elena Tognini Bonelli (Università di Lecce / The Tuscan Word Centre, Italy)
This paper presents a case, through argument and example, for the establishment of a new discipline within linguistics, and within corpus linguistics. The provisional name of Corpus-driven Linguistics (CDL) is offered in order to point the contrast with what we refer to as “corpus-based linguistics”.
It will be argued that corpus-based work relates corpus data to existing descriptive categories adding a probabilistic extension to theoretical parameters which are already received, i.e. established without reference to corpus evidence. On the other hand, corpus-driven work attempts to define the categories of description step by step, in the presence of specific evidence from the corpus.
The difference between the two approaches, however, is not only methodological but also qualitative. The corpus-driven approach at times leads the analyst to question some of the most basic received wisdom about language, and often points to new determining elements which have to be incorporated in the description. The strict interrelation between grammatical and lexical choices shown up by corpus evidence is a case in point and leads the corpus-driven analyst to postulate new categories.
While corpus-based work tends to focus on paradigmatic nodes, corpus-driven work highlights the importance of syntagmatic multi-word units made up of lexical, grammatical, semantic elements which, together, perform a specific function within the context they operate in; at the paradigmatic level, they represent a single choice. What emerges in corpus-driven work is a new unit of currency of description.
The paper will provide a theoretical discussion of the issues related to a corpus-driven approach and will illustrate them with practical examples. It will argue for the validity of the approach and recommend its adoption as one of the future challenges for the discipline of corpus linguistics.
Corpus-driven linguistics (CDL): position statement
One question we should ask ourselves is: what are the basic requirements of a new discipline, to differentiate it from those nearby? We will argue for the establishment of such a new discipline starting from the assumption that these are:
• a set of goals toward which the research hopes to move, in careful stages.
• a philosophical standpoint, an orientation to the data that is not as well developed elsewhere.
• a unique, or at least particular methodology.
• a set of theoretical and descriptive categories for articulating the content of the research.
• an accumulating body of knowledge that would be difficult if not impossible to acquire from other sources (though it may be confirmed or questioned by alternative approaches).
These assumptions are discussed in turn and some examples are presented to illustrate the argument.
The primary goal of CDL is to make exhaustive and explicit connections between the occurrence and distribution of language items in text, and the meanings created by the text. There are two principal issues here:
1. Texts are physical objects and meanings are unobservable, so claims could be made to justify a direct association between formal and functional elements. The safeguard here is the intuition of the language user, who must in some sense be satisfied that the connections offer an illuminating explanation of the way language text creates meaning.
2. This goal is not unique, in that every adequate theory of language might be seen to adopt an identical goal. That does not invalidate the distinctiveness of the discipline, indeed it confirms that it is in the mainstream of linguistics. But the precise phrasing of the goal is not already adopted by other linguistic theories, so there are aspects of emphasis and priority that may still serve to give CDL a distinctive flavour.
CDL considers axiomatic the statement that meaning arises as much because of the combination of choices in a text as because of the individual contribution made by the meaning of each choice. The combinations of choices are recognised as relevant to meaning in several areas of language patterning, but not in the comprehensive way that is implied in the above statement. Grammatical structure deals with combinations, and their relative sequencing, but not the specific choices of linguistic items - only linguistic abstractions like classes, elements of structure, etc. The study of idiom, and the recent flowering of phraseology, are concerned with specific combinations, but idioms are seen as occasional events, and phraseology is not tightly associated with meaning creation.
In CDL, the approach to language patterning is holistic. Any step away from the physical data is taken with care, and regarded as a weakening of the description unless compensated for by much greater generalisation.
The essential methodology of CDL is to exercise the researcher's intuition in the presence of as much relevant data as can be assembled. It is accepted that there is no such thing as a theory-neutral stance, but in CDL the attempt is made to suppress all received theories, axioms and precepts and to rely on the standpoint above to guide the initial stages of any investigation. Obviously, as experience grows there will be new hypotheses that arise from the investigations, and if those are generally accepted they will form part of CDL methodology.
Specifically in the present intellectual climate, CDL does not accept prima facie those theories, axioms and precepts that were formulated before corpus data became available. These are not rejected or dismissed - the accumulated insights of centuries of research are not to be put aside lightly - but they are to be re-examined in the new frameworks where, instead of the scholar having to struggle to gather a sufficient amount of data, (s)he has now a plethora of data at his/her disposal.
CDL is not immediately concerned with positioning itself vis-à-vis the tradition of theoretical and descriptive linguistics. The results of research probes so far (e.g. Tognini Bonelli 2001) show convincingly that there are substantial differences between the patterns that are discovered in language corpora and those that are anticipated by the mainstream linguistic work of the last century, and especially of the last half-century. CDL does not, therefore, accept the agendas that are popular in other branches of linguistics, but will pursue its own goals probably for some time, until the theoretical position is more fully articulated and the descriptive system is elaborated.
For some time this may seem a laborious way of working, since so much of language structure seems to be non-contentious, but the methodology of CDL requires different standards of attestation from other approaches, and joins other strict sciences in expecting that all results are replicable.
As the main lines of description become clear, it is to be expected that a descriptive apparatus will take shape in response to the descriptive needs. Some basic categories are already postulated, clustering round the central concept of a “functionally complete unit of meaning” (Tognini Bonelli 1996, 2001).
Body of knowledge
The awareness that there was special knowledge to be gained from a corpus, not available from any other source, and certainly not from unaided introspection, was the original impetus to establish CDL as a separate branch of linguistics. This knowledge has accumulated over years in several centres, and an ad hoc terminology has grown up around it because it could not be described with the normal apparatus of linguistics.
Gunnel Tottie & Hans Martin Lehmann (University of Zurich, Switzerland)
As is nowadays considered non-standard as a relativizer when used as in (1):
(1) Well I know one person as’ll eat it. (Biber et al. 1999: 609)
However, as can be used after antecedents containing same and such in Standard English. An instance is shown in (2), where as is in variation with other relativizers, that and zero.
(2a) John always bought the same car as Mary did.
(2b) John always bought the same car that Mary did.
(2c) John always bought the same car Ø Mary did.
Notice that (3) is ungrammatical without same:
(3) *John always bought the car as Mary did
This type of variation has received little attention in the literature on relativization. We intend to report on the occurrence of as and its equivalents in current spoken and written British and American English from BNC and the Longman Corpus, respectively, as well as in British and American newspaper corpora from the nineteen-nineties. In this paper we will restrict ourselves to the use of as after same.
The corpora chosen for this study represent both spoken and written registers of American and British English. We chose LSAC (Longman Spoken American Corpus) and the spoken component of BNC (from here on BNC-S) for spoken language. Written language is represented by the 1999 issues of The [London] Times (TLN) and The Los Angeles Times (LATM).
Relative constructions with same are too infrequent for manual retrieval. For the purpose of locating these constructions we therefore used techniques described in Lehmann (1997a; 1997b; In press) and Tottie and Lehmann (1999). At the heart of this approach is the search for surface patterns of word-class sequences, as this is crucial for locating elements that are absent in the surface structure, such as zero relativizers.
For the present study we used the engcg-2 tagger (cf. www.conexor.fi) to annotate the corpora with word-class information. The retrieval process used here is based on the general patterns for relative constructions with the additional requirement that the antecedent must begin with the same. Since the retrieval patterns will include the whole antecedent, they will have to implement an NP model. The one used for this study can deal with most premodification, e.g. as in (4). Postmodification is limited to of + NP, as in (5).
(4) British television is merely seeing the same ratings-led drift that American television went through in the 1960s. (TLN955269538)
(5) Her motionless figure and the aloof and even tones of her voice in the scene of her summoning Siegmund to Wallhalla and the electric change of tone in the first moment of her pity showed the same grasp of the implications of character which has made her Isolde a thing of exceptional power. (TLN954581415)
In terms of functions of the relativizer the present study covers the full range from subject (6)–(8), direct object (9)–(10), to oblique relatives as in (11)–(12). (Cf. Keenan & Comrie 1977; 1979). As a consequence it covers the relativizers who, whom, which, that as well as prepositional relatives with either pied-piping or stranding, and the alternatives where when, why and the marginal how. As mentioned above this study is limited to antecedents which start with the same.
(6) If people are gonna sit around the house while they're unemployed, they're probably the same people who sit around the house when they're on a Saturday and Sunday when they're working you know. (BNC:HEN)
(7) And you will probably be aware that at the consultation draft stage, which shows the same boundaries as are in the deposit plan, […] (BNC:FNM)
(8) […] , it is the same person who was born in the, in the manger at Bethlehem, he is now, after having died, been raised […] (BNC:J8Y)
(9) […] and in Maryland we had a situation that kind of evolved into the same kind of political row Ø you would expect when a company loses a long time business . (BNC:HE6)
(10) But our department hasn't changed, the women are just doing the same job as they did sixty years ago. (BNC:H03)
(11) The season before that, they had the misfortune to visit the City Ground when Nottingham Forest, languishing in the same parlous state in which they find themselves today, had just appointed Stuart Pearce as caretaker manager. (TLN956399874)
(12) “…If the wind keeps on blowing from the same direction Ø it's coming from today, there will be some big numbers out there,” he said. (TLN953335977)
Table 1. Relativizers after same in BNC-S and The [London] Times.
Biber, D., Johansson, S., Leech, G., Conrad, S. and Finegan, E. (1999) Longman Grammar of Spoken and Written English. Longman, Harlow.
Keenan, E. and Bernard C. (1977) Noun Phrase Accessibility and Universal Grammar, Linguistic Inquiry 8: 63-99.
Keenan, E. and Bernard C. (1979) Data on the Noun Phrase Accessibility Hierarchy, Language 55: 333-351.
Lehmann, H. M. (1997a) Things Nobody Can See or Hear: Automatic retrieval of Zero Elements in a computerised Corpus. Unpublished MA Thesis. University of Zurich.
Lehmann, H. M. (1997b) Automatic Retrieval of Zero Elements in a Computerised Corpus. In Ljung, M. (ed.) Corpus-based Studies in English. Rodopi, Amsterdam, pp. 179-194.
Lehmann, H. M. (In press) Zero Subject Relative Constructions in American and British English. In Peters, P. (ed.) Proceedings from ICAME 2000. Rodopi, Amsterdam.
Joe Trotta & Mats Johansson (Halmstad University, Sweden)
In the unmarked word order in English, a premodifying AdjP itself embedded in an NP occupies a slot between the head noun and the determiner, i.e. it is normally a string which, for the sake of simplicity, can be abbreviated as Det + AdjP + N as in a big problem, the industrious student, his faithful companion, etc. In certain circumstances, however, the AdjP must precede the determiner, resulting in a AdjP + Det + N construction:
1. a. It wasn’t that big a problem. (cf *It wasn’t a that big problem)
b. b. How short a time we had for our visit!) a ? (cf *A how short time we had…)
c. c. He felt it wasn’t so foolish an idea at all. (cf *He felt it wasn’t an so foolish idea…)
The unusual position of the adjectives big, short and foolish in (1), is sometimes described as a consequence of the fact that they themselves are modified by a so-called ‘intensifiers’ (here that, how, and so), a combination which requires a ‘preposing’ of the AdjP to a position preceding the determiner (see Seppänen 1978; Quirk et al. 1985: 834-835; Delsing 1993: 138-146).
However, this same word order can also be noted in another similar construction involving how in which the AdjP is not premodified by an ‘intensifying’ item but rather by an ordinary interrogative (Trotta 2001: 43-45):
2. a. How big a problem is it?
b. How large an area did it cover?
Moreover, in addition to the structures noted in (1) and (2), there is another variation on this string which has been generally overlooked or ignored in descriptive grammars of English:
3. a. How big of a problem do you think that is? (CDC: npr/07)
b. b. […] then […] the causes of the anxiety and the depression er are no longer that big of a problem and the person can handle it then. (CDC: ukspok/04)
c. c. I live in too big of a house (Abney 1987: 325)
d. [… ] lets not go over how stupid of a move this was, I already know […] (online: http://woozle.org/netforum/myjournal/a/8--126.96.36.199.1)
Given this background, the aims of the present paper are twofold:
i) We examine and describe the ignored variation of the AdjP + Det + N construction exemplified in (3). We investigate its frequency in the major corpora as well as look into its use in a selected study of internet material. Aside from the quantitative study of this string, previous analyses of the AdjP + Det + N construction are scrutinized and we consider the relation between the types noted in (1), (2) and (3).
ii) ii) With the help of our case study, we approach some of the problems faced by the corpus linguist in dealing with low-frequency phenomena such as: What kinds of conclusions can one draw about the ‘naturalness’ of a construction based on it’s frequency in a particular corpus? How does the composition of a particular corpus affect the way a researcher views a low-frequency construction? Is the size of a corpus the most relevant issue and if so, how big of a corpus is a big enough corpus?
Abney, S. (1987) The English Noun Phrase in its Sentential Aspect. Unpublished PhD Dissertation, MIT.
Delsing, L-O. (1993) The Internal Structure of Noun Phrases in the Scandinavian Languages. Department of Scandinavian Languages, University of Lund, Lund.
Quirk, R., Greenbaum, S., Leech, G. and Svartvik, J. (1985) A Comprehensive Grammar of the English Language. Longman, London.
Seppänen, A. (1978) Some notes on the construction “adjective + A + noun”, English Studies 59: 523-537
Trotta, J. (2001) Wh-clauses in English: Aspects of Theory and Description, Rodopi, Amsterdam & Atlanta.
Åke Viberg (Uppsala University, Sweden)
This paper is one in a series of papers dealing with verbs of possession from a crosslinguistic perspective. The polysemy of the rather language-specific Swedish possession verb få ‘get;may’ has been treated in an earlier study (Viberg, forthcoming). This study will be concerned with two other basic verbs within the field, namely ge ‘give’ and ta ‘take’. In addition, a brief sketch will be made of the field of possession verbs based on the representation in the project Swedish WordNet (related to EuroWordNet). The analysis of give and take is based on translation corpora: a restricted pilot corpus consisting of extracts from novels in Swedish with translations into English, German, French and Finnish and the complete set of occurrences of Swedish ge and ta and English give and take in the English Swedish Parallel Corpus (ESPC) prepared by Altenberg & Aijmer.
Intertranslatability. The Swedish possession verb få treated in an earlier study turned out to be very language-specific with respect to its semantic patterning. Its closest equivalent in English get was used as a translation of only 11.7% of the 2043 occurences of få in the Swedish original texts in the ESPC. The verbs ge/give and ta/take have patterns of polysemy which are rather similar in Swedish and English and this appears to reflect universal tendencies (see Newman 1996 for ‘give’). Swedish ta and ge were both translated with their closest English equivalents (give and take, respectively) in around 43% of the cases in the ESPC and this is a relativly high proportion for highly frequent verbs with many meanings. In the presentation, examples will also be given of translational equivalents of individual meanings of these highly polysemous verbs.
Patterns of polysemy. The basic aim of the paper is to account for the patterns of polysemy of ‘give’ and ‘take’. Below some examples are given of the prototypical and some of the extended meanings of ge and ta.
1. ge ‘give’
Prototypical meaning: Possession
- Vi tänker ge henne eget rum. MF
'We're thinking of giving her a room of her own,' she went on.
"Wir haben vor, ihr ein eigenes Zimmer zu geben.
-Nous envisageons de lui donner une chambre individuelle.
-Olemme ajatelleet antaa hänelle oman huoneen.
Abstract possession: Mental
De nygifta ger intryck av burget självmedvetande: IB
The newly weds give the impression of well-to-do self-assurance:
Die Frischvermählten machen den Eindruck wohlhabenden Selbstbewußtseins:
Les nouveaux mariés donnent une impression d'assurance aisée:
Vastanaineista saa mielikuvan hyvässä asemassa olevien ihmisten itsetietoisuudesta:
och försökte undvika att trampa på något som gav ljud ifrån sig. KE
, trying to avoid treading on anything that might make a noise.
und vermied tunlichst, auf etwas zu treten, was ein Geräusch machte
en évitant instinctivement de faire du bruit ou de marcher sur des choses qui pourraient en faire.
ja yrittäen välttää potkaisemasta mitään mistä lähtisi ääntä.
Mia gav sig iväg efter samtalet. KE
Mia left after the conversation.
Mia ging nach dem Gespräch.
Mia partit après le coup de fil.
Mia lähti heti puhelun jälkeen.
Indirect causation (Finnish only)
Jag lutade mig över min dockteater, lät ridån vällustigt höja sig /---/ IB
I leant over my toy theatre, letting the curtain rise voluptuously
Ich beugte mich über mein Puppentheater, ließ den Vorhang wollüstig /---/ hochgehen.
Penché sur mon théâtre de poupées je laissais voluptueusement se lever le rideau
Minä kumarruin nukketeatterini ylle, annoin esiripun nautinnollisesti nousta
In the example where ge is used in its prototypical meaning, the primary equivalent appears in all the translations. In the various types of extended meaning, the variation is much greater in spite of the fact that it is often possible to find parallel cases where the primary equivalent is used. The patterns of polysemy are similar at a general level but there are many cases of language-specific more or less lexicalized phrasal combinations. However, certain extended meanings are relatively language-specific at a general level. In Swedish, Subject-centered motion is an example of this. The use of Finnish antaa ‘give’ as an indirect causative has no parallel among the other four languages in the pilot corpus but has parallels in several other languages (e.g. Chinese, Luo).
2. ta ‘take’
Ta fyra praliner, men se dig för, så att du inte blir ertappad. IB
Take four chocolate creams, but mind they don't catch you'.
Nimm vier Pralinen, aber sieh dich vor, daß man dich nicht erwischt.”
Prends quatre pralines, mais fais bien attention qu'on ne te surprenne pas.
Ota neljä karamellia, mutta pidä varasi, ettet jää kiinni.
Han rufsade honom i håret. Tog i honom. KE
The man rufffled his hair and touched him.
Er zauste ihm durchs Haar. Faßte ihn an.
Le gars lui ébouriffa les cheveux. Le toucha.
Mies pörrötti hänen tukkaansa. Kosketti häntä.
Jag hämtade mor som tagit sig till teatern genom snöovädret. IB
I went to fetch my mother, who had made her way to the theatre through the snowstorm
Ich holte meine Mutter ab, die sich durch den Schneesturm bis zum Theater vorgekämpft hatte.
Je suis allé chercher ma mère qui venait d'affronter la tempête de neige pour venir au théâtre. [COME]
Kävin noutamassa äidin, joka oli tullut teatteriin lumimyrskyn läpi. [COME]
Newman, J. (1996) Give. A cognitive linguistic study. Mouton de Gruyter, Berlin.
Viberg, Å. (forthcoming) Polysemy and disambiguation cues across languages. The case of Swedish få and English get. To appear in: Granger, S. and Altenberg, B. (eds) Lexis in Contrast. Benjamins, Amsterdam.
Anne Wichmann (University of Central Lancashire, United Kingdom)
Work on attitudinal intonation to date has mainly been based on intuition and anecdote. This is not in itself a problem and does not, of course, preclude a subsequent, more quantitative approach. However, the step from intuition to large-scale corpus studies is in the case of attitudinal intonation not so simple. First of all, intonation patterns with 'marked' attitudinal implications are frequent. Secondly, their identification is subjective. Thirdly, as with all discourse phenomena, they are much harder to annotate and to search for. It may be that this is an area where corpora simply are no longer useful, and that any quantitative analysis will simply require a large quantity of qualitatively analysed examples, collected fortuitously and selected by intuition. Before conceding in this way, I should like, however, to consider ways in which existing (and future) corpora might still offer more systematic ways of studying the elusive phenomenon which is 'attitudinal intonation'.
What can corpora do?
Past attempts to classify the 'attitudes' which intonation can convey has led in the literature to a plethora of labels, both positive and negative (from rude, angry and condescending to cheerful, friendly and comforting). Each reference to 'attitude' brings new labels, but there is little evidence of an underlying system. Psychologists (e.g. Scherer) and linguists (Couper-Kuhlen, Wichmann) have attempted to bring some order to these endlessly proliferating lists, by identifying different kinds of 'attitude' and 'affect' in human interaction. As yet, however, no-one seems to have investigated to what extent such labels are actually used by the participants in interaction, and if so, which they use most frequently. I shall report the results of a preliminary search of this kind.
Finding 'potential' speech acts
Assuming access to a grammatically annotated corpus such as the ICE GB, it is possible to search for certain kinds of constructions that one might expect to constitute speech acts with the potential to convey 'attitude'. A recent example is a study of Wh-questions in ICE GB (Wichmann & Cauldwell 2001). Lexical searches can also reveal certain speech acts such as thanking, apologising, requesting (see Aijmer 1996), all of which can, of course, be conveyed in a neutral way but also have the potential to be said with a marked 'tone of voice'. Unfortunately, until corpora are labelled pragmatically, studies of this type are restricted to what can be achieved by indirect lexical or grammatical searches.
Finding the 'norm'
There is another way in which corpora can guide us towards, if not point directly to, attitudes. I have argued elsewhere (Wichmann 2000a, b) that interpersonal attitude needs to be discussed within a pragmatic framework. To perceive a voice as 'friendly' is to make an inference based on a combination of textual, prosodic and contextual information. Intonational 'attitudes' are implicatures or inferences based on available information and arise particularly when there is some kind of mismatch between prosody and text, or prosody and context. Such a mismatch or incongruity signals: 'do not take this at face value; seek another meaning'.
The observation of incongruity inevitably relies on our knowledge, implicit or otherwise, of what is congruous. This is where corpora can be of assistance - in uncovering what is 'normal'. With this information we can then identify incongruous or abnormal behaviour, which may then be the key to explaining perceived 'attitude'. In this way, any work which manages to uncover prosodic patterns that reliably co-occur with certain speech acts, or in certain contexts, contributes to the study of attitude (Aijmer 1996 is a good example).
Where corpora fail
While current interest in spoken language is strong, the attention paid to the primary data, namely the sound files themselves, varies immensely. The BNC, for example, has not made the recordings readily available. ICE GB, on the other hand, has been innovative in aligning the sound with the text, and the Corpus of Spoken Dutch consists of sound files and orthographic transcriptions. What we as corpus users cannot expect any longer is to be provided with a prosodic annotation. The painstaking work of annotating the Spoken English Corpus (SEC) or the London-Lund Corpus (LLC) is a thing of the past. For those whose expertise lies elsewhere, but who would like to use prosodic information in their analysis, this is a considerable loss.
Even if it were available, auditory prosodic analysis is not enough. Neither linguists nor the speech community can now rely exclusively on auditory annotation. There are important prosodic phenomena which can only be explored instrumentally, especially global prosodic parameters such as pitch range, loudness and speech rate. Intonational phonology also explores its categories both in terms of meaningful auditory distinctions and in terms of their acoustic reality. This requirement for instrumental methods of analysis has important technical consequences for recording. However authentic the conversation, if it has been recorded in a busy pub it is useless for instrumental analysis. The first requirement of speech corpora is therefore that the sound should be treated as essential primary data. This means that availability of sound files is a pre-requisite. In addition, the sound needs to be of good enough quality for instrumental analysis; more thought should therefore be given to ways of collecting high-quality recordings.
Secondly, while selection criteria that distinguish between private vs. public conversation, scripted vs. unscripted, monologue vs. dialogue, have ensured a wide range of speaking styles, the resulting corpora are still too homogeneous for interesting work on prosody. The interaction represented in current corpora is symmetrical in terms of power relationships, co-operative, and also for the most part affectively neutral. For studying attitudinal intonation we need more asymmetrical, uncooperative, and confrontational discourse.
The most useful work on prosody is now being done outside the corpus linguistic community by two very different groups: speech technologists and conversation analysts. Conversation Analysts carry out minutely detailed work on small amounts of data. In my view, they tend to work with a fairly atheoretical, impressionistic view of prosody, ignoring both past and recent developments in intonational phonology. They also work at such a level of detail that systematic generalisations cannot be made. However, these minute observations at least raise useful hypotheses which could in theory be tested quantitatively. Prosody is also an important a focus of research in the speech community, with the aim of modelling prosody for use in applications such as automatic dialogue systems. High-quality recordings and very sophisticated analysis techniques are used; the sophistication of the approach has furthered our knowledge not only of intonational phonology, but also of more global prosodic characteristics of speech which could not be identified impressionistically but which reveal how these characteristics function in interaction. Unfortunately the kind of data being used is extremely limited - specially elicited conversations in laboratory conditions, mostly replicating some kind of goal-directed service encounter.
The corpus linguistic community could play an important role by mediating between these two extremes - encouraging quantitative analysis but on the basis of linguistically interesting data. As it is, corpus linguists have little to offer those whose interest lies in the sounds of speech.
Aijmer, K (1996) Conversational Routines. Longman, London.
Scherer, KR. (forthcoming) Psychological models of emotion. To appear in Borod, J. (ed.) The neuropsychology of emotion. Oxford University Press, New York.
Couper-Kuhlen, E. (1986) An Introduction to English Prosody. Arnold.
Wichmann, A. (2000a) Intonation in Text and Discourse. Pearson Education, London.
Wichmann, A. (2000b) The attitudinal effects of prosody and how they relate to emotion. Proceedings of ISCA Workshop on Speech & Emotion, Newcastle, Northern Ireland pp 143-147.
Wichmann, A. and Cauldwell, R. (2001) Wh Questions and attitude: the effect of context. Presented at CL2001, Lancaster.
Chris Allen (University of Halmstad, Sweden)
This poster reports on the initial stages of a corpus-driven study of causality in modern English. The study builds on recent work on sublanguages and on small scale ‘local grammars’ written to parse narrowly-defined communicative functions. To date complete or partial local grammars have been written to analyse full-sentence dictionary definitions, evaluation, causality and duration. Proponents of local grammars argue that existing grammatical analyses and terminology - whether structurally or functionally-motivated - may be too general to be useful in the parsing of such narrow communicative functions. A local grammar in contrast can be thought of as a series of ad hoc grammatical categories designed to analyse a specific semantic or communicative notion. Local grammar parses have the further advantage of being more semantically transparent than their general language counterparts, raising the prospect that they could be used in future information retrieval/extraction applications. It is envisaged that the parsing of unrestricted text can then be carried out using an array of individual local grammars. On the basis of the corpus evidence being analysed I hope to be able to identify the most significant lexicogrammatical patterns through which cause and effect relationships are realised which can then serve as the basis for a small scale grammar of causality at a later date.
Niek Brom, Inge de Mönnink & Nelleke Oostdijk (University of Nijmegen, The Netherlands)
The research conducted by Biber (1988; 1990; 1995) in which he uses the MF/MD method for the automatic classification of texts has been criticized by a number of people, incl. Oostdijk (1988), Altenberg (1989), and most recently Lee (2000). The criticisms that are put forward concern various aspects of the method and the way in which it has been applied. Among these are the size and the nature of the corpus and the samples used, and the selection of grammatical features. It has also been suggested that the dimensions as postulated by Biber (1990) are not as pronounced as he claims. One of the questions that has so far remained unanswered is the following: would it have made a difference if Biber had used a different set of grammatical features? At the time that Biber conducted his research, no corpora were available that had been annotated with detailed syntactic information. Since then, however, the fully parsed ICE-GB corpus (Nelson 1996) has become available. It is against this background that we decided to conduct a study which aimed to clarify the issue as to whether the data set has any effect on the results obtained in applying the MF/MD method. In the poster presentation the results of this study are presented.
Description of the experiments
The present study includes three experiments. In these experiments, the set of linguistic features is varied, while the corpus data is kept the same. The three sets of linguistic features are as follows:
1. 1. Biber's set of 67 variables
2. 2. a set of 129 tags
3. 3. a set of 103 sentence structures
The corpus which is used is the 1-million word ICE-GB corpus. It consists of 500 texts of approximately 2000 words each.
In the first experiment, Biber's application of the MF/MD method as described in Biber (1988) was copied as meticulously as possible on the ICE-GB corpus. While making use of the syntactic annotation available in the ICE-GB corpus, every attempt was made to stay as close as possible to Biber's 67 grammatical features. In other words, we converted Biber's algorithms into Fuzzy Tree Fragments. By means of this experiment we wanted to investigate whether the classification as postulated by Biber would hold when the research was replicated on a different corpus. In the attempt to copy the linguistic features, some obscurities, inconsistencies and shortcomings in the algorithms used by Biber came to light. The availability of syntactic annotation provided the opportunity to search for most linguistic features using only one (complex) fuzzy tree fragment and to improve on some of the original search schemes. Only changes that improved the precision and recall of the original algorithms were carried through.
For the second experiment, the full set of 332 tags found in the ICE-GB corpus was reduced to make it suitable for factor analysis. The reduction was established by ignoring some of the word classes (e.g. pause, punc, interjec) and features (e.g. comp, disc, procl), ignoring incomplete tags (e.g. V(ditr), N(sing)), and ignoring all features for some major word classes (e.g. ADJ, NUM).
As input for the third experiment we took all sentence patterns with a frequency of occurrence of 70 or higher. To establish this set, we looked at the functions of the daughters of the highest node in the sentence. This results in a total of 4815 different functional structures of which only 103 occur 70 times or more (with a highest frequency of 5390 for the structure subject-verb-subject complement).
These last two experiments differ from the first in the fact that the set of linguistic features is not based on previous micro-analysis. While this may complicate or even render impossible the functional interpretation of linguistic features in terms of an underlying dimensional structure, it has the clear advantage that no prior research into the communicative functions of features is needed to carry out the MF/MD analysis. If it turns out that a factor analysis on tags and/or sentence structures still results in a meaningful text classification, this method can be used for the automatic categorization of texts. Automatic text categorization in turn has its applications in areas such as information retrieval and data warehousing and in determining the representativeness of corpus design.
Description of the results
In the first experiment, the (normalized) frequency counts resulting from the fuzzy tree fragments were used as input for the factor analysis. The resulting factorial structure consists of five factors, instead of Biber's seven factors. When comparing both factorial structures, we find some differences, but also some striking similarities. For example, the 13 variables that have a salient positive loading on Factor 1 in our factorial structure also load on Biber's first factor and the variables in our Factor 3 all load on Biber's Factor 2. Some of the differences that we found can be explained by the use of the improved search algorithms. The feature 'wh-relative clause on subject position', for example, was improved to include cases where the head of the noun phrase is not realized by a common noun, or where the postmodifying relative clause is preceded by a prepositional phrase. In Biber's calculations this variable has a positive loading on Factor 3, while in our factorial structure it has a high negative loading on Factor 1.
If we compare the mean factor scores of the ICE-GB genres with Biber's scores, we find that Biber's first dimension 'involved versus informational' is clearly reflected in the mean scores for our Factor 1, with private and public conversations on one end of the scale, and academic writing on the other. Biber's second dimension 'narrative versus non-narrative' is reflected in the scores for our Factor 3, with creative writing on one end of the scale, and instructional writing on the other. The distribution of scores for our Factor 2 are not directly reflected in Biber's study, but it seems to make a clear distinction between spontaneous speech on the one hand, and scripted speech and writing on the other.
The results from our second and third experiment cannot be compared with Biber's results directly, since we are dealing with both a different corpus and a different set of linguistic features, but we can compare the mean factor scores for the genres with the results of our first experiment. On doing so, we find that the results for the set of tags are very similar to those of the first experiment. Again, we find a distinction between involved vs. dimensional, narrative vs. non-narrative, and written vs. spoken. While approximately the same distinctions are found for the set of structures, the mean factor scores are here far less distinct.
In the current study Biber's classification (cf. Biber 1988) was largely reproduced, using the same linguistic features. It was shown that the availability of syntactic annotation simplified and improved the search for the linguistic features considerably, effecting the classification on some points. At the same time, however, it was shown that a factor analysis carried out on the frequency counts of only a set of tags resulted in largely the same text categories. Assuming that the categorization is a useful one, this would indicate that the automatic categorization of tagged texts is feasible.
The current study also brought to light some limiting conditions on and suggestions for the successful implementation of the MF/MD method. With regard to the design of the corpus, it can be concluded that the number of texts in some of the 32 genres in ICE-GB is too small to obtain significant differences in factor scores (using the Anova test). For a successful implementation of the MF/MD method the number of texts per genre should be increased. Another improvement of the MF/MD method can be found in the availability of a more accurately annotated corpus. The syntactic analyses found in ICE-GB still show numerous mistakes and inconsistencies. Here the use of tags as input for the factor analysis provides a solution, since accuracy scores for tagging are much higher than for parsing.
On the whole it can be concluded that, while Biber's factorial structure was largely reproduced, the text classification for English is not yet stable and can still be improved bearing in mind the findings of the present study.
Altenberg, B. (1989) Review of ‘Variation across speech and writing’ by D. Biber (1988), Studia Linguistica 43(2): 167-174.
Biber, D. (1988) Variation across speech and writing. Cambridge University Press, Cambridge.
Biber, D. (1990) Methodological issues regarding corpus-based analysis of linguistic variation, Literary and linguistic computing 5: 257-269.
Biber, D. (1995) Dimensions of register variation. Cambridge University Press, Cambridge.
Lee, D. (2000) Unpublished Ph.D. Thesis. Lancaster University, Lancaster.
Nelson, G. (1996). The Design of the Corpus. In Greenbaum, S. (ed.), 27-35.
Oostdijk, N. (1988) A corpus linguistic approach to linguistic variation, Literary and linguistic computing 3: 12-25.
Andreas Eriksson (Göteborg University, Sweden)
It remains unclear, however, why there exists such an amazing variety of ways to express these concepts and why tense and aspect distinctions generally constitute the most difficult part of the language system for non-native language learners, even if the target language is genetically very close to the native one. (Vet & Vetters 1994: 1)
The tense, mood and aspect (TMA) systems of English and Swedish are in many respects similar and it should therefore be possible to use the same basic framework in order to describe the systems. Despite the similarities, however, many Swedish students have struggled with the TMA categories while trying to improve their English. When learners reach an advanced level, they will make fewer tense mistakes, but the system is still likely to cause difficulties1. Problems which may arise are, for instance, what tenses to use in a certain text type and what tenses to use in a particular part of a text. The difficulties will probably not be describable only in terms of misuse, but also as matters of over- and underuse.
The first aim of the present study is to give a description of the usage of tense, mood and aspect in argumentative essays written by advanced Swedish learners and to compare it with the usage found in essays written by English-speaking students. Part of this aim is thus to locate areas in the TMA system which might cause difficulties for learners. The second aim is to scrutinize what TMA categories are used in particular parts of a text. Thirdly, special attention will be paid to so-called TMA shifts, e.g. shifts from the present to the past tense, or from the past to the present perfect. The study thus goes beyond the sentence level and views tense from a text linguistic perspective. Granger (1999: 196-197) has pointed out that a major problem for learners seems to be that they adopt a clause- or sentence-level approach when selecting tenses. Furthermore, it is emphasised by Granger that tense plays an important role in textual cohesion (Granger 1999: 198). Consequently, there seem to be sound reasons for studying learners' and native speakers' tense usage within the scope of text linguistics.
The material is collected from two corpora consisting of argumentative essays: the Swedish component of the International Corpus of Learner English (SWICLE) and the Louvain Corpus of Native English Essays (LOCNESS). Other material will also be used in order to see if there is any difference in terms of the use of TMA categories between argumentative texts written by learners (both native and non-native speakers) and professional writers.
Model for the investigation of TMA categories
I am currently trying to work out a model which links English and Swedish tense, mood and aspect systems, both in terms of form and function. At this point, this work builds on a model by Nordlander (1997), where tense, mood and aspect are seen as expressing three types of time by means of three binary distinctions:
Factual time - realis-irrealis: mood
Internal time - perfective-imperfective: aspect2
External time - anterior-non-anterior: tense
The idea is that three values, one from each pair, are always expressed in a situation, either by explicit marking or implicitly by the absence of marking (cf. the notion of obligatory sets in Bybee & Dahl (1989). The difference between sentences like The Swedish people made a mistake and The Swedish people will make a mistake is one of tense and mood, since the former is ‘realis, perfective, anterior’, whereas the other is ‘irrealis, perfective, non-anterior’. The model makes it possible to compare what forms learners and native-speakers use to express, e.g. irrealis mood. The importance of this type of analysis was indicated in a minor case study carried out on four texts from each corpus, since, in the case of conditionality, learners more often used open conditions, while native speakers used hypothetical conditions, as exemplified in (1) and (2):
(1) As it is now, I think it may be a mistake if Sweden does not join the union. (SWICLE)
(2) If a single Europe should lead to British troops fighting on behalf of Europe instead of its own nation, most sovereignty would be lost (LOCNESS: esee30).
The model will be elaborated in order to capture more fine-grained distinctions, e.g. different degrees of hypotheticality.
Another area that will be looked into is the complexity of the verb phrase, in order to see if, as one perhaps would expect, greater complexity is found in the texts of native speakers. Measures of complexity are yet to be looked into, but one example of verb phrases with high complexity are phrases containing infinitival complements as in He is believed to have killed the dog.
Bybee, J. L. and Dahl, Ö. (1989) The Creation of Tense and Aspect Systems in the Languages of the World, Studies in Language 13: 51-103.
Granger, S. (1999) Use of Tenses by Advanced EFL Learners: Evidence from an Error-tagged Computer Corpus. In Hasselgård, H. and Oksefjell, S. (eds) Out of Corpora. Studies in Honour of Stig Johansson. Language and Computers: Studies in Practical Linguistics 26. Rodopi, Amsterdam, pp. 191-202.
Nordlander, J. (1997) Towards a Semantics of Linguistic Time. Swedish Science Press, Uppsala.
Vet, C. and Vetters, C (eds) (1985) Tense and Aspect in Discourse. Trends in Lingustics. Studies and Monographs 75. Mouton de Gruyter, Berlin/New York.
Monika Hägglund (Göteborg University, Sweden)
The aim of the thesis is to investigate the use of English verb-particle constructions (VPCs) in advanced Swedish learners’ written (and spoken) interlanguage. This will include what is traditionally referred to as phrasal verbs and also phrasal-prepositional verbs (Quirk et al. 1985).
The written material consists of argumentative texts from the Swedish component of the International Corpus of Learner English (SWICLE). A comparison will be made with a native speaker norm e.g. the Louvain Corpus of Native English Essays (LOCNESS). If time allows, the thesis will be expanded to include spoken language e.g. from the Louvain Database of Spoken English Interlanguage (LINDSEI)).
I am interested in analyzing general differences between the two speaker groups as well as register differences. Do Swedish advanced learners use different VPCs from native speakers? Is their use appropriately adapted to the register?
Since the Swedish learners are advanced, it is unlikely that differences can be solely accounted for by misuse. Instead one may anticipate instances of over- and underuse. Both over- and underuse may be due to influence from the mother tongue since Swedish also allows for VPCs.
The fact that both languages show this common typological characteristic does not mean that the combinations are identical, nor does it mean that the acquisition of English VPCs is unproblematic for Swedish learners. Previous research has shown that Swedish learners prefer transparent combinations to opaque combinations (Sjöholm 1995) in English, i.e. what one may refer to as prototypical usage.
The aim of the actual analysis is to encompass both semantics and syntax. When looking at the semantics of the VPCs a cognitive perspective might be fruitful. The general assumption in that theoretical framework is that the meaning of these constructions is not arbitrary. Instead the semantics can be explained by the components, in literal as well as in metaphorical and metonymic use (Lindner 1982, Morgan 1997, Hampe 1997). In other words, a particle, such as out can be can be metaphorically extended to encompass non-animate, non-physical and non-spatial entities (Johnson 1987), such as to stay out of trouble or to figure something out. This gives rise to the following question:
Extensions: Do NNSs use the same kind of metaphoric/metonymic extensions of the particles/and the verbs as NSs do?
Another issue concerns the choices available to the NSs and the NNSs. While finish may be said to compete with finish off, many VPCs have a different one-word synonym, usually of Romance origin. Such is the case with take off and depart, for example, which leads to the following question:
Synonyms: What differences are there between NSs and NNSs regarding the use of VPCs vs. potential one-word synonyms?
Other types of variation, of a syntactic kind, which may be interesting to study from a contrastive perspective, have to do with object placement and the possible optionality of the particle:
Object placement: Is there a preference for medial or final placement of the object by the two speaker groups (in cases where there is a choice)?
Optional particles: Is there a difference between the NSs and the NNSs concerning the use of particles that are not part of the valency of the verb, e.g. finish off vs. finish?
On a deeper level, the thesis aims to account for the differences and similarities between the NS and the NNS groups, e.g. in terms of transfer, avoidance strategies and proficiency.
Beate Hampe, J. (1997) Toward a Solution of the Phrasal Verb Puzzle: Considerations on some Scattered Pieces, Lexicology 3(2).
Berman, R.A. and Slobin, D.I. (1994) Relating Events in Narrative: a Crosslinguistic Developmental Study. Lawrence Erlbaum Associates, Hillsdale.
Biber, D., Johansson, S., Leech, G., Conrad, S. and Finegan, E. (1999) Longman Grammar of Spoken and Written English. Longman, London.
Johnson, M (1987) The Body in the Mind. The Bodily Basis of Meaning, Imagination, and Reason. University of Chicago Press, Chicago.
Lindner, S. (1982) What Goes Up Doesn’t Necessarily Come Down: The Ins and Outs of Opposites, Papers from the Regional Meetings, Chicago Linguistics Society 18: 305-23.
Morgan, P. S. (1997) Figuring Out figure out: Metaphor and the Semantics of the English Verb-Particle Construction, Cognitive Linguistics 8: 327-57.
Sjöholm, K. (1995) The Influence of Crosslinguistic, Semantic and Input Factors on the Acquisition of English Phrasal Verbs. A Comparison between Finnish and Swedish Learners at Intermediate and Advanced Level. Åbo Academi University Press, Åbo.
Quirk, R., Greenbaum, S., Leech, G. and Svartvik, J. (1985) A Comprehensive Grammar of the English Language. Longman, London.
Tomoko Kaneko (Showa Women’s University, Japan)
This is a report on a pilot study on some of the Japanese data compiled for the Louvain International Database of Spoken English Interlanguage (LINDSEI) project. Data for the Japanese portion of LINDSEI was collected and reached its goal of 50 samples in May 2000. We are still adding more data. One of the purposes of the study is to find possible ways of using international corpus data like LINDSEI to investigate learners’ interlanguage. However, to serve as a pilot study, the present study is based only on a part of the Japanese data. While transcribing the data, the researcher noticed that the subjects had difficulty in sustaining the past tense frame in their speeches. The researcher wanted to know to what extent the Japanese learners correctly used different types of past tense verb forms and to speculate on why they could not sustain the past tense frame in their speaking. Thus, the research questions are as follows:
1. 1. To what extent do learners use the past tense forms correctly?
2. 2. What kind of qualitative features are shown in learners’ errors in the past tense verbs?
Thirty out of fifty LINDSEI Japanese files were checked and tagged for obligatory context for the past tense forms of regular, irregular, be-, and auxiliary verbs. WordSmith analysis tool was utilized to analyse the quality and quantity of the target language forms.
For all four types of verbs, correct use slightly out-numbered incorrect use. The accuracy order was 1) irregular verbs, 57.8%, 2) regular verbs, 53.2%, and 3) be-verbs and auxiliary verbs, 50.0%, which suggests that the subjects had acquired irregular verbs earlier than regular verbs. However, since the learners used various types of verbs in regular verbs, while they used only limited types of irregular verbs, the concept of accuracy order judged by obligatory context should be questioned. Nevertheless, at the very least, the result of the present study suggests that verbs that are simpler in form are not necessarily easier for learners to acquire.
From the comparison of the frequency word lists of the correct and incorrect regular verbs, it was suggested that the learners recognized the need to mark past tense more easily with verbs which show objective facts than with verbs which show states of mind. The finding is similar to those of Bardovi-Harling and Reynolds (1995), who found that learners seem to find it easier to mark past tense when referring to completed actions than when referring to states and activities which may last for extended periods. This kind of issue also needs to be investigated further from a cognitive linguistic point of view. The comparison of the frequency word list of the correct and incorrect irregular verbs shows that the amount of input seems to be a crucial factor in the use of the past tense forms, and that past tense seems to be marked more easily at the beginning of a long utterance than in the middle or toward the end of the utterance. It may be that the task to convey the meaning requires so much concentration that the learners aren’t able to concentrate on the form that is known to be correct. Levelt (1989) reports that in the case of native speakers, vocabulary errors and the speakers correcting them toward the end of their utterances are rare. This contrast between native speakers and learners also needs to be investigated in later studies. Finally, it was suggested that one reason for the fact that the learners had difficulty in using were, the plural form of the be-verb, seems to relate to its pronunciation. Were needs more effort to pronounce than was, maybe because of the increased value of the vowel.
Cluster and collocation lists show that learners often self-repaired their verb utterances, especially when they didn’t have confidence in the form and/or pronunciation. The lists also show that chunk-learning in classrooms worked well for Could you ~ ? but didn’t work for I wanted to ~ . It is obvious from our experience that Japanese learners rarely practise I wanted to ~ pattern in classrooms, while they often practise Could you ~? as a form for requesting. It is also interesting to note that past tense was marked more easily when there was another marker of time frame in the utterance, for example, the conjunction when. The forty-five cases of correct use of the cluster when I was support this idea. Because of the limit of the total number of tokens in the files, it was not possible in the present study to find other conjunctions that seem to work as markers of time frame. This point also needs to be studied in the future.
Despite the small amount of data, we could gain some valuable insights into aspects of learner language. We believe that further study using more files, especially the comparative study using data from other language backgrounds, will give us more insights into language acquisition by non-native learners.
Bardovi-Harling, K. and Reynolds, D. (1995) The role of lexical aspect in the acquisition of tense and aspect, TESOL Quarterly 29(1): 107-31.
Levelt, W. J. M. (1989) Speaking: From intention to articulation. The MIT Press, Boston.
Alexander Kautzsch (University of Regensburg, Germany)
The majority of studies of non-finite verb forms in English tend to be selective to some extent. That is, researchers often restrict themselves to an analysis of one non-finite verb form, e.g. the infinitive (e.g. Fischer 1997, Fanego 1992), or the ing-form (Moessner 1997); or of one certain syntactic function of non-finite constructions, e.g. adverbial clauses (Ljung 1997) or verbal complementation (Los 1998). One exception is Svartvik and Quirk (1970), who study a wide variety of functions of non-finite clauses but restrict themselves to Chaucer. This tendency to study a certain period or a certain author is the second type of selection frequently encountered. Of course, these limitations are necessary to provide in-depth analyses of the respective feature or period under scrutiny.
Nevertheless, a more general approach might have its benefits for our understanding of the changing shape of English and the mechanisms involved in linguistic change. This is why the present study differs a little from common practice: it sets out to investigate if and how the English language has changed over the last 500 years as far as the syntactic behaviour of all non-finite verb forms is concerned.
In order to do so, I will provide an empirical analysis of the non-finite verb forms (i.e. plain infinitive, to-infinitive, for-to-infinitive, past participle and ing-form) of eight high frequency verbs in English (bring, come, tell, know, play, write, help, and drink), based on the Helsinki, Brown, and LOB corpora. [For the selection of these verbs I used Hofland and Johansson’s (1982) word frequencies. However, the most frequent ones – go and make – were set aside due to their high degree of semantic variability and change through time.]
A general survey of the total frequencies from Old English to present-day English (PDE) shows very interesting developments towards a more equally distributed occurrence of non-finite verb forms. That is, it appears that the plain infinitive heavily loses ground through time, while the to-infinitive, the past participle, and — even more so – the ing-form continually rise in frequency from Old English to the present. As a result, in present-day English the frequency differences between the non-finite verb forms have become smaller than they used to be.
More detailed analyses are capable of showing which syntactic categories are responsible for this change in the appearance of English through time. Since, as we all know, change usually comes after variation, it is obvious that those grammatical environments in which one non-finite form is the (almost-)categorical one — as for example the past participle in the passive, the ing-form in the continuous form, or the plain infinitive after auxiliaries — should be set aside. But those contexts in which more than one non-finite verb form is possible give interesting insights in changes though time.
These ‘crucial’ grammatical categories are non-finites as verbal complements with and without subject, adverbial clauses, modification, nominal clauses. It can be shown, for example, that the to-infinitive gains a vary large share in these contexts through time. It increases steadily as verbal complement with subject from Old English to the present and as verbal complement without subject from Old English to late Early Modern English, from where it drops a little towards the present. Further, it is a frequent modifier through time but decreases in this function after Early Modern English. In adverbial clauses the to-infinitive remains a relatively constant number two, while in nominal clauses it is the almost categorical number one from Early Middle English onwards. I think this short glimpse into the results neatly demonstrates the power of this approach.
In addition, this presentation is the initial stage of a project that intends to document non-finite verb forms in standard varieties of today's English. Since part of the analysis is based on the Brown and LOB corpora, first results of differences between American and British English — to be precise: between written American and British English of 1961 — can already be displayed. Similar to the overall survey, the interesting contexts are also the ones in which non-finites appear variably. And it appears, indeed, as if American and British English have different preferences for the usage of non-finite verb forms. Verbal complementation by means of non-finites is just a case in point: without subject, British English and American English tend to favour the to-infinitive and the ing-form, respectively, whereas in complementations including a subject for the non-finite form, the results are reversed.
On the whole, I think this long-term approach to the development of non-finite verb forms is a very fertile one for two reasons. First, by providing an integrated analysis of all non-finites it becomes possible to receive an overall impression of the changing shape of a relatively large part of the English grammar through time. Second, on a more general level this presentation shows that it is possible to watch syntax change. The only prerequisite is to use a corpus with a relatively big temporal range which in this case is achieved by means of a combination of diachronic and synchronic corpora.
Fanago, T. (1994) Infinitive marking in Early Modern English. In Fernandez, F., Fuster, M. and Calvo, J. J. (eds) Historical Linguistics 1992. Benjamins, Amsterdam, pp. 191-203.
Fischer, O. (1997) Infinitive marking in Late Middle English. Transitivity and changes in the English system of case. In Fisiak, J. and Winter, W. (eds) (1997), pp. 109-34.
Fisiak, J. and Winter, W. (eds) (1997) Studies in Middle English Linguistics. Mouton de Gruyter, Berlin.
Hofland, K. and Johannsson, S. (1982) Word Frequencies in British and American English. Norwegian Computing Centre for the Humanities, Bergen.
Ljung, M. (1997) A genre-based study of English subordinator-headed non-finite and verbless adverbial clauses. In Nevalainen, T. and Kahlas, T-L. (eds) To Explain the Present: Studies in the Changing English Language in Honour of Matti Rissanen. Societe Neopholologique, Helsinki, pp. 375-94.
Los, B. (1998) The rise of the to-infinitive as verb complement, English Language and Linguistics 2(1): 1-36.
Moessner, L. (1997) -ing constructions in Middle English. In Fisiak, J. and Winter, W. (eds) (1997), pp. 335-49.
Svartvik, J. and Quirk, R. (1970) Types and uses of non-finite clauses in Chaucer, English Studies 51: 393-411.
The Corpus of Early English Correspondence (CEEC), reaching over almost 300 years of personal letters and including 2.7 million words, is now being expanded. This diachronic corpus designed by the Sociolinguistics and Language History project at the English Department of the University of Helsinki is being augmented to cover the 18th century as well as the last two decades of the 17th century.
The corpus, which now contains around 6,000 letters from almost 800 informants, and its published sampler version (CEECS), of roughly half a million words no longer in copyright, serve as material for the research group interested in applying modern sociolinguistic methods to historical linguistics. The team has produced well over 50 publications. A collection of pilot studies can be found in Nevalainen and Raumolin-Brunberg (eds) (1996).
When completed, the CEEC supplement will cover personal correspondence to the end of the 18th century. The letters are selected from published collections of correspondence often by comparing several different editions in order to provide the corpus with accurate extralinguistic information. The supplement will be provided with a same kind of sender database as created for the CEEC. It will give researchers easy access to different sociolinguistic variables including the writer's provenance, social status, sex, education, age and relationship between the writer and the addressee. The text level and parameter coding of the supplement will also follow those used in the CEEC.
The aim of the team is to include letters not only from the higher social ranks of English society but to include material written by people from all walks of life. It is also in our interest to try to include as much material from female informants as possible. At this stage of the compilation process it could be said that there seems to be a large number of published collections of private correspondence covering the 18th century. It should be noted, however, that some writers of the period, especially literary men and women, often addressed their letters either to a large circle of people or intended them to be published later. Therefore, one of the challenges in the process is to find material that fulfils the criterion of private correspondence.
The CEEC supplement now contains two types of material. Firstly, new letter collections have been added to the corpus to cover the last two decades of the 17th century, since the CEEC covers the time period until 1681. In addition to that, some of the letter collections that are already included in the corpus have been complemented. Secondly, according to the original purpose of the supplement, new private correspondence collections, mainly from the first half of the 18th century, have been added to the corpus. Put together, this means that the supplement now consists of about 350,000 words.
Keränen, J. (1998) The Corpus of Early English Correspondence: Progress Report. In Renouf, A. (ed.) Explorations in Corpus Linguistics. Language and Computers: Studies in Practical Linguistics 23. Rodopi, Amsterdam, pp. 29-37.
Nevalainen, T. and Raumolin-Brunberg, H. (eds) (1996) Sociolinguistics and Language History. Studies Based on the Corpus of Early English Correspondence. Language and Computers: Studies in Practical Linguistics 15. Rodopi, Amsterdam.
Ilka Mindt (Universität Würzburg, Germany)
This poster aims at presenting a quantitative analysis of the frequency and the realisation of clause elements (e.g. subject, direct object etc.) and phrases (e.g. noun phrase, adverb phrase etc.) in ICE-GB.
Four aspects will be dealt with:
a) the frequency of clause elements (e.g. subject, object, adverbial);
b) the frequency of phrases (e.g. noun phrase, adjective phrase);
c) the function of phrases (e.g. noun phrases functioning as subjects, direct objects etc);
d) the realisation of clause elements (e.g. subjects being realised as noun phrases and clauses)
The unit of analysis is the top-level of sentences. In a sentence such as:
However, the important thing is that we are having a house warming party [...]
the top-level taken into account for the analysis looks as follows:
the important thing
that we are having a house warming party
a) The frequency of clause elements
The most frequent clause elements are: verbs, subjects, adverbials, discourse markers, direct objects, and subject complements. These six elements account for 94.2% of all clause elements.
b) The frequency of phrases
Clause elements are realised by phrases or by clauses. Thus, the frequency of phrases and of clauses functioning as clause elements is considered. Clause elements are most frequently realised by noun phrases, verb phrases, clauses, prepositional phrases, and adverb phrases. These five types account for 96.6% of all phrases.
c) The function of phrases
Next, the function of phrases is considered. Each phrase is investigated independently. The different function of phrases as clause elements will be presented. Additionally, the relative frequency of each functional realisation is given.
Noun phrases can function as subjects (69.7%), as direct object (18.5%), as subject complements (8%), as adverbials (2.1%) and as other clause elements (1.7%). Each phrase type will be dealt with in turn.
d) The realisation of clause elements
Finally, the reverse view is taken. The focus will be on the phrasal realisation of clause elements. Example: subjects are realised in 96.8% of all cases as noun phrases and in 3.2% as clauses. Each clause element will be dealt with in turn.
Vladimira Miňovská (UJEP Usti nad Labem, Czech Republic)
Our students have been contributing to the ICLE for several years. Before we started our attempts at analysing the corpus material, the students seemed quite confident about their essays as all they worried about was grammar. The pioneering analysts began to discover, as it has gradually turned out, surprising facts for both the students and the teachers.
These included the school-born myths:
- - a text is a good quality text if it follows the rules of grammar and cohesion
- - writing is a creative activity that cannot be learnt/taught
- - writing in a foreign language is more or less just writing following different rules of grammar and using (English) equivalents of Czech words
- - students learn to write by reading
- - students can write in their mother tongue.
Very little attention is paid to writing in general in our country. Students not only feel helpless but they are used to it and they accept it. Teachers do not feel very confident either, and correcting students’ writing is usually to a large degree a question of grammar only.
Student analyses have so far covered the areas of prepositions, sentence connectors, sentence length, writer/reader visibility, tag frequencies, punctuation, modal auxiliaries, definite articles, passives, infinitives and participles.
Frequency counts based on the Czech subcorpus (CZ) and LOCNESS suggested significant overuse of among in CZ. One of the underused prepositions was between. These two prepositions have only one equivalent in Czech – mezi. It seemed that speakers of Czech are not able to apply what they learn about the use of between and among. The corpus material revealed that the overuse was primarily caused by translating from Czech and producing constructions that do not exist in English. The misuse was a result of ‘correct’ application of simple rules of among/between usage presented by textbooks and materials used in our schools.
The underused connectors in CZ ICLE could have been easily predicted. They are the long, ‘difficult’ words we all know, but not well enough to risk using them. Overused connectors mostly belong to informal registers and are typical of spoken language. The student’s conclusion that the overuse in written texts is caused by the recent obsession with the highly misunderstood concept of communicative approaches at schools is, probably, correct.
Sentence length was compared in Spanish, French, Finnish, Dutch and Czech subcorpora and LOCNESS. The distribution curves showed three groups. LOCNESS with the longest sentences, the CZ subcorpus with the shortest sentences and all the other subcorpora in the middle group. The student-researcher concluded that the findings reflect the fact that the Czech students at the time of their essay writing were not as advanced as students in other countries as English was widely introduced into our schools only recently.
Inspired by Stephanie Petch-Tyson (1998), who claims that the pronoun I appears in ‘chains’, the student wrote his own simple programme to extract a map of occurrences of I in the Czech corpus. It shows that the high frequency of the word indicates that there are writers prone to overusing I rather than that all users use the pronoun too extensively. His findings do not support Petch-Tyson’s conclusions that the pronoun is connected with verbs in the past tense, often used for recounting personal experiences. In the Czech essays past tense verbs were not very frequent.
Existential there shows significant overuse. As the Finnish overuse is even more prominent, it seems that there may be used to help solve some problems of English word order to students with highly flective mother tongues. Noun frequency brings the Czech students nearer to imaginative, informal prose than the informative prose of argumentative essays, which relates to a high frequency of verbs as well. Underuse of prepositions in CZ seems a natural result of students’ avoiding constructions where they are likely to make mistakes.
Czech students like all other students, (except Finns) overuse punctuation. They tend to apply the Czech rules – commas preceding all dependent clauses.
The Czech subcorpus has the verb frequencies even higher than other learner corpora, again except the Finnish. It puts the essays even more clearly in imaginative texts. The second reason is the fact that CZ has a higher number of sentences per 1,000 words.
Overuse of can and underuse of may. A more thorough analysis would probably prove that students use can instead of risking may in extrinsic modality. The Czech equivalent of may expresses only intrinsic modality. Have to is slightly preferred to must. Textbooks and popular grammars used in our country stress probably too much that must is personal and should not be used for facts. Students know that they should avoid personal references in essays, so they avoid must. Will and would frequencies are different in different subcorpora. The author came to the conclusion that the reasons relate to the frequency of the topics. CZ favourite topics are No 1,6,11, while in the French corpus No 12 seemed to be the favourite.
Definite and indefinite articles
There are no articles in Czech. The Czech learners use about 25% less articles than native speakers and less articles than other non-native speakers whose languages also have no articles. A striking gap between Finnish and Czech learners led to the question of noun frequencies, which turned out to be very similar. The student’s solution presents an interesting learner perspective: input. Generally our input is very limited (all films are dubbed), while the Finns are exposed to English every day. They do not have to rely on learning how English articles are used but they can hear them being used.
Passives, Infinitives, Participles
Czech students use less passives than any other group, which supports the findings in previous chapters about the CZ writing being less formal. With its high frequencies of infinitives, the CZ subcorpus displays the “speech-like” quality (Granger 1998) of our students’ writing. Czech students’ score of participle frequencies is only a fraction better than French.
Students’ papers based on the ICLE material have pointed out interesting facts, but first of all they have introduced the concept of interlanguage. Interlanguage awareness development based on the ICLE gives a new perspective to our English teacher education courses.
Lene Nordrum (Göteborg University, Sweden)
My thesis will be a contrastive study of nominal and verbal style in English, Norwegian and Swedish expository prose. Nominal style is defined by the use of long, elaborate noun phrases and nominalizations, and verbal style by the use of coordinated or subordinated clauses containing a finite verb. Nominal style is frequently commented upon in studies on the difference between spoken and written language. Here, it is emphasized that heavy use of nominal forms (nouns, nominalizations and gerunds) is the most salient trademark of written language, whereas a style characterized by low lexical density and a high proportion of finite clauses is a feature of spoken language (Biber 1995, Biber et al 1999, Chafe 1985, Halliday 1994). Moreover, in the Longman Grammar of Spoken and Written English (Biber et al 1999), it is found that the registers academic writing and news are considerably more nominal than fiction and conversation.
In this connection, it seems that register theory in Systemic Functional Grammar (SFG), and particularly the notion of grammatical metaphor, provide a fruitful approach to the register differences associated with nominal and verbal style. In SFG it is claimed that the basic grammatical pattern in English is clausal (Halliday 1994). Halliday holds that it is not until language users are capable of expressing themselves through clauses that they are ready for the slightly more sophisticated task of reconstruing propositional content into nominal categories. To exemplify: a child would not say in times of engine failure, rather he would use the clause whenever the engine fails (Halliday 1994: 353). When the content of a clause is restructured into a phrase, SFG uses the notion grammatical metaphor. Nominalizations, then, are non-congruent realizations of language, as opposed to their verbal counterparts which are congruent. Halliday & Martin (1993: 238) argue that a congruent piece of discourse consists of structures which maintain a natural relation between semantic and grammatical categories. That is, when the child first learns his mother tongue, he comes to the understanding that objects and things are denoted by nouns whereas processes are denoted by verbs, and this comes to be the logic on which he builds his grammar. A nominalization, then, is a grammatical metaphor, since a process is naturally denoted by a verb. In effect, grammatical metaphor provides us with a useful distinction between nominal and verbal style, in that nominal style might be described as non-congruent and verbal style as congruent language use.
The concept of grammatical metaphor, then, provides me with a useful theoretical notion to contrast nominal and verbal style in my material. The following research questions form the foundation of the study:
Regarding the question whether comparable registers are equally nominal/verbal in the three languages studied, a small pilot study on one English text, A History of God from Abraham to the Present the 4000-year Quest of God by Karen Armstrong, and its Norwegian and Swedish translations suggests that Norwegian might be less nominal than both English and Swedish. This can be exemplified by the example below. Put in SFG terminology, the Swedish translation (i), takes over the grammatical metaphor in the English original, whereas the Norwegian translation (iii) renders the grammatical metaphor the placing congruently as a verbal participle (set) in a postmodifying non-finite clause (grammatical metaphor underlined):
The placing of this incident in stark juxtaposition to the awesome revelation (...)
Placeringen av händelsen i skarp motsättning till den vördnadsbjudande uppenbarelsen (...)
Denne hendelsen stilt opp slik i skarp kontrast til den fryktinngytende åpenbaringen (...)
[This incident set up in this manner in sharp contrast to the awesome revelation (...)]
Also, connected to the second research question posed above, Halliday & Martin’s (1993) findings that grammatical metaphor serves different purposes in different text types raise the question whether the function of grammatical metaphor across registers is similar language-internally as well as between languages.
The texts from the pilot study were taken from the non-fiction part of the two sister corpora, the English-Norwegian Parallel Corpus (ENPC) and the English-Swedish Parallel Corpus (ESPC). These corpora provide the main bulk of material for my thesis, since they share many of the same English originals and thereby facilitate a comparison of translated texts. Texts from the British National Corpus (BNC) and Swedish and Norwegian monolingual corpora will be added to the material from ENPC and ESPC. The total amount of texts included in the study remains to be fixed.
To sum up, then, I find that some of the theoretical assumptions made in SFG provide me with a useful starting point for a contrastive study on nominal and verbal style in English, Norwegian and Swedish expository prose.
Biber, D. (1995) Dimensions of Rregister variation: a Cross-linguistic Comparison. Cambridge University Press, Cambridge.
Biber, D., Johansson, S., Leech, G., Conrad, S., Finegan, E. (1999) Longman Grammar of Spoken and Written English. Longman, London.
Chafe, W (1982) ‘Integration and Involvement in Speaking and, Writing, and Oral Literature’. In Tannen, D. (ed.) Spoken and Written Language: Exploring Orality and Literacy. Norwood, New York.
Halliday, M.A.K. and Martin, J.R. (1993) Writing Science. Literacy and Discursive Power. Falmer Press, London.
Halliday, M.A.K. (1994) An Introduction to Funtional Grammar. 2nd edition. Arnold, London/New York/Sydney/Auckland.
Ravelli, L.J. (1988) Grammatical Metaphor: an Initial Analysis. In Steiner & Veltman (eds), Pragmatics, Discourse and Text. Pinter Publishers, London.
Pam Peters & Adam Smith (Macquarie University, Australia)
The EDOC project aims to address this fundamental gap by developing a corpus of digital documents (i.e. ones designed in the first instance to be read on screen), collecting two major nonfictional genres: the instructional and the informational (or the procedural and the expository, in Longacre's 1974 taxonomy of discourse types). These two categories were selected because of their importance in public communication, and as functions for which many feel that electronic delivery is well suited. They are also genres which commonly appear in printed form, and therefore readily compared on aspects such as structure and segmentation. Computerized samples of these print-based equivalents are already held in the Macquarie ACE and ICE corpora, and therefore readily matched and subjected to computational analysis.
The EDOC corpus of electronic documents contains more than 100,000 words of instructional and of informative prose, extracted from about 100 sources, in samples whose finished length varied (from 200 to 8,000 words). The texts, all of Australian origin, so as to match ACE and ICE selections on that parameter, were selected from bibliographies of online documents, and with the aid of Yahoo search categories, limited by region. The sources were checked to ensure that they were indeed "digitally born", not merely electronic copies of preexisting paper-based documents. Whole texts were sampled because of our interest in overall structure, and the very question as to what communicators thought was a reasonable length for screen reading. The instructional texts embraced several subcategories, including those of:
1. 1. teaching structured information
2. 2. regulating behavior
3. 3. constructing something
4. 4. solving a problem
Informational texts were subcategorized in terms of their readerships (general and specialized), and discourse type: as news reports, thematic or argumentative articles, or scientific, in the broad sense.
Within this body of data, the project aims to compare textual structure and segmentation on several levels, using both conventional cues, such as the writer's recourse to section and paragraph divisions, as well as the segmentation of text into lists and individual sentences. Sentence punctuation is therefore of interest, as well as the use of headings marked by contrasts of font and line space. The actual font changes and spaces may of course reflect the reader's browser settings rather than the author's selections; so here and elsewhere the analysis is confined to those aspects of text structure which cannot be adjusted by the browser, but are of the author's making. We can nevertheless take advantage of whatever information is provided in HTML coding or any XML style sheet associated with the document, as a basis for annotating the structural and segmental features.
In pilot work to establish the parameters of analysis, a set of four texts in each of the categories of instructional and informational prose was analyzed in terms of its structural components from the level of text down to sentence or subsentence (e.g. listed item). The texts were chosen from the middle range of length in each category, the average instructional text being about 1,200 words long, and 2,100 words long for informational. The results except for the whole discourse, were then compared with samples from the appropriate categories in ACE and ICE. The hypothesis was that the documents from EDOC would be segmented into shorter units at all structural levels. This proved to be true for the instructional samples, but not so consistently for the academic ones, some of which seemed to take little account of the impact of reading longish documents on screen.
It seems that writers-to-screen do often anticipate the finite bounds of the screen, and its effect on screen readers. This is presumably an intuitive response, given that there is so far relatively little received wisdom in the field.
Hale, C. (1996) Wired Style (HardWired).
Longacre, R. (1974) An anatomy of speech notions (Peter de Ridder).
Caren Sanders (University of Zurich, Switzerland)
The ZEN Corpus (Zurich English Newspaper Corpus) was compiled at the University of Zurich. The corpus covers the years from 1671 to 1791 and contains a selection of early English newspapers published in London. The project will be finished in early summer 2001 and a published version of the corpus is expected to be available in a CD-format. The Corpus consists of about 1,500,000 words at the moment and provides a suitable tool for analysing newspaper language of this period.
It is known from contemporary newspapers that there are sections such as foreign or domestic news, sports, and business classified by the publisher. However, regarding 17th and 18th century news publications these divisions are not yet consistently indicated, because the articles are mere flat texts, frequently starting with a place- and dateline, followed by the news report. Additionally, the publications are news collections rather than chronologically and systematically ordered items.
The purpose of the presentation is to find out more about the structures of 17th and 18th century newspaper language. The starting point was the clear and classifiable category of texts type advertisements. What became obvious from a very early stage was that the term advertisement was used in a fundamentally different way 300 years ago. In those times there were advertisements for things lost and stolen, people who had run away or were needed as well as announcements. It was impossible to handle all of these categories at the same time, so we concentrated exclusively on product advertising.
The next step was to strip the ZEN Corpus and to store all advertisements and especially all medical advertisements in a subcorpus. Medical advertisements were chosen because there was a manageable number of them (in contrast to book advertisements, for example,) so that we could be sure to have a clear overview of the area of research. The medical advertisement subcorpus contains 341 advertisements which corresponds to a total of about 87,573 words; these very specialised texts formed the beginning of our research.
The main aim was to examine the results of the distribution of medical advertisements per decade. The result was as expected: there are more advertisements towards the end of the 18th century in comparison to the beginning of the 17th century. Furthermore, the content structure and the average number of words per advertisement were investigated. We struggled with the increase of the average number of words per advertisement towards the end of the period of the ZEN Corpus but found out that advertisements in German newspapers had undergone a similar process at that time. Another point which seemed worth looking at were names of products and also the most frequent adjectives in premodifying position which describe medical products.
Anna-Brita Stenström (University of Bergen, Norway)
Why collect a corpus of London teenage talk? One answer to this question is that English teenage talk - and teenage language in general - had been very little investigated at the beginning of the 1990s, and no ‘large’ corpus of English teenage talk was available for research. A second answer is that new trends that appear in the London teenage talk can be expected to have a great potential for influencing not only teenage language in general but also adult language.
The compilation of COLT (The Bergen Corpus of London Teenage Language) was modelled on that of the BNC (The British National Corpus) in May and September 1993 and amounts to roughly half a million words. The 13 to 17 year-old male and female recruits, who recorded the conversations they engaged in with their peers, came from five socially different London school districts (Barnet, Camden, Hackney, Tower Hamlets and Hertfordshire). The tape recordings have been orthographically transcribed and word-class tagged. Both versions are available for research on the Internet, and a first CD-ROM version appeared early this year. Finally, a COLT book (Trends in Teenage Talk: Corpus compilation, analysis and findings) will, hopefully, be published by Benjamins later this year.
Is teenage talk synonymous with ‘bad language’? That is of course a matter of opinion, but many of the features that struck us as typically used by the COLT teenagers are often heavily criticized by adults, e.g. the use of slang, vague words, tags, old words with new meanings, and not least the rich use of swearwords. A sociolinguistic comparison of some of the findings can be summed up as follows:
§ § Slang: the older boys and girls use more slang than the younger ones. Overall, the boys use slang more frequently than the girls and also more ‘dirty’ slang words. The use of slang, including dirty slang, dominates in Tower Hamlets, which is the school borough that is lowest on the social scale, whereas ‘proper’ slang dominates in Hertfordshire, which is highest on the social scale
§ § Swearing: The boys swear more than the girls and use more of the stronger ones (fuck, shit, bloody). Among the boys, it is the youngest ones that swear most, compared to the older ones among the girls. As regards school borough, swearwords are more common in Tower Hamlets than anywhere else.
§ § Tags: A comparison of the use of okay, right, yeah and innit shows that innit totally dominates speaking followed by okay, right and yeah. With the exception of okay, the use of the tags increases with age. With respect to gender, the boys are more frequent users of especially yeah, while the girls use innit more often. As to school borough, okay dominates in Camden, right in Barnet, yeah in Camden and innit in Hackney.
More of this with illustrations on the poster.
Ann Taylor, Anthony Warner, Susan Pintzuk & Frank Beths (University of York, United Kingdom)
The York-Helsinki Parsed Corpus of Old English (YCOE) is a project currently underway at the University of York to create a 1.5 million word, part-of-speech tagged and syntactically parsed corpus of Old English. This corpus is fully compatible with the recently released Penn-Helsinki Parsed Corpus of Middle English, second edition (Kroch and Taylor), the York-Helsinki Parsed Corpus of Old English Poetry (Pintzuk and Plug), and the comparable Penn-Helsinki Parsed Corpus of Early Modern English (Kroch and Santorini) currently being developed at the University of Pennsylvania. The end result of these linked projects will be a coherent series of annotated electronic corpora for the major historical periods of the English language.
The YCOE is parsed with a limited hierarchical system of labelled parentheses. Each token consists of one main clause along with its associated subordinate clauses bracketed and labelled according to type, and is referenced by text, page and token number. Within the clause, words are labelled by part of speech, and phrasal constituents are bracketed and labelled by formal type (NP, PP, etc.) with additional functional information (such as subordinate clause type, predicate, etc.) added in some cases. A limited number of empty categories, such as traces of wh-movement and various types of extraposition and scrambling, have been included to make semantic relations clear. Our goal in designing the system was to balance the need for extensive and accurate morphological and syntactic information with a reasonable rate of progress in producing usable output. In addition we have tried to avoid introducing annotations where the judgments of grammarians are known to differ or to be unstable.
The data contained in the YCOE and its sister corpora are accessed by means of a specially designed search engine, CorpusSearch, which provides all matching tokens along with summary statistics in response to a simple query. The search engine is highly customizable and even includes a coding function, the output of which can be used as input to statistical packages such as varbrule, SPSS, DataDesk, etc. The kinds of data, both qualitative and quantitative, that can be extracted quickly and easily from such corpora provide an unparalleled new resource for empirical studies in the history of the English language.
Shunji Yamazaki (Daito Bunka University, Japan)
This poster is part of a large-scale study of adjectives and adjectival collocations in the Wellington Corpus of Written New Zealand with the aim of comparing its diversity with the Brown and LOB Corpora. In this poster I describe some syntactic characteristics of adjectives in three corpora of written English (Brown, LOB, and Wellington). The distribution across genres of attributive and predicative adjectives differs considerably. For example, academic writings and, within press material, reviews, show high use of attributive adjectives (I bought a new cap the other day), whereas predicative adjectives (John will be able to come tomorrow) are particularly frequent in romantic fiction.
Comparing specific attributive adjectives in the three corpora reveals distinctive differences between the corpora. A comparison of frequency lists shows for example a higher frequency of insular or inward-looking adjectives in Brown (American, national, local, federal, military, central, religious), as against more international adjectives in Wellington (international, Australian, British, French, English, foreign, European, overseas). These findings are consistent with differences in the various countries’ political and economic power and self-sufficiency.
In the poster the recently collected Italian component of the International Corpus of Learner English (ICLE) will be presented and taken as the starting point for a corpus-based investigation. Among the mistakes made by advanced Italian learners of English the frequent confusion between even if and even though to express hypothetical or concessive meanings respectively will be considered. This appears to be due to interference with Italian where the same form (anche se) can be used to express both meanings.
In order to explore the issue more extensively, statements will be drawn from recent descriptions of English (e.g. Quirk et al. 1985 and Biber et al. 1999), pedagogical grammars, various types of dictionary, native English learner corpora (e.g. LOCNESS), English corpora, native speaker judgements, and the other ICLE subcorpora. Comparison with these different sources of information will show whether this confusion is a typical Italian mistake, a widespread learner mistake or, indeed, a very subtle conceptual distinction which is gradually disappearing in language use.
Michael Barlow (Rice University, USA)
Bilingual or parallel corpora provide, in effect, an accumulated store of numerous translation choices. We can thus use aligned bilingual corpora to explore the notion of translation equivalence and, more generally, the relationship between the form-meaning links of one language and the form-meaning links of a second language.
One way to extract the information inherent in such corpora is by using a parallel concordance program such as ParaConc. Like other concordance programs, ParaConc facilitates research into the lexical, syntactic, and semantic patterns of a language. It differs from other programs in that it is designed to work with parallel texts, i.e., texts in two languages that are translations and are aligned, typically sentence by sentence in most cases. To distinguish the two languages when necessary, I will refer to them in the abstract as languages A and B. It is also possible to think of the software more broadly as a way of searching parallel representations of what might be labelled as "the same thing." For example, each tier in a tiered linguistic representation or alternative writing representations may be reworked as parallel texts.
For ParaConc to work correctly, the two parallel texts must be aligned. The current version of the program does not include an aligner and so the alignment must either be carried out manually or accomplished using a separate program or utility. Each sentence (or segment) must either be followed by a paragraph break (or some other special character) or be identified using beginning and ending tags.
The use of the program typically involves the following stages. Once the parallel texts have been loaded, the investigator specifies a morpheme, word, or phrase (i.e. a search term) from language A (or language B), and the program then finds and displays all the instances of the search term. The concordance results are typically displayed in KWIC format, i.e. with the search word in the centre of the window, along with a context of preceding and following words. In addition, the program shows in a second, lower window all the sentences in the second text from language B that contain the meaning associated with the search term. Such is the general situation, but naturally it sometimes happens that some item or some meaning in the source text does not appear in the target text, at least not in the segment that is displayed.
At this stage the user can click on lines of interest from language A (possibly sorted 1st right, 1st left etc.) and observe the corresponding lines from language B. It is also possible to search for, and highlight, words in language B, which leads to a situation in which there are parallel search words. In this case, the collocates of both search terms can be obtained and sorting can be applied to language A or language B. In addition, there is a "hot words" utility which attempts to suggest language B translations for the language A search term.
A variety of features commonly found with monolingual concordancers are present in ParaConc and will be demonstrated as time permits. A history of different versions of ParaConc and access to different beta test versions of the program is provided on the web at www.ruf.rice.edu/~barlow/parac.html.
Estelle Dagneaux, Sylviane Granger, Fanny Meunier, Stephanie Petch-Tyson & Xavier Vilret (Université catholique de Louvain, Belgium)
The International Corpus of Learner English currently contains over 2 million words of writing by EFL learners from a variety of mother tongue backgrounds. The corpus has the following defining characteristics:
Ø Ø homogeneity: it contains a well-specified type of data (academic, mainly argumentative texts of c. 500 words each in their integral form) from advanced EFL learners
Ø Ø documentation: detailed information relating both to the learner and to the task is stored alongside the texts
The corpus lends itself to two types of research:
Ø Ø theoretical SLA (Second Language Acquisition) research: being a very controlled corpus, ICLE provides an ideal platform for assessing the role played by transfer in SLA
Ø Ø applied ELT research: ICLE provides a solid empirical foundation for describing advanced learners' interlanguage and on that basis for producing more focused and hence more efficient ELT tools (textbooks, grammars, dictionaries).
The first CD-Rom version of the corpus will be available in late 2001. It will also be accessible online via a Java web interface designed both to query and manage the database. An authentification procedure will govern access to the system according to user’s rights. The database itself is a relational database (powered by postgreSQL) which will assure quick response times.
Knut Hofland (University of Bergen, Norway)
The Bergen Corpus of London Teenage Language (COLT) is the first, and so far the only, existing corpus of English teenage language talk that is available for research worldwide. The Corpus is now available on a set of 3 CD-ROMs with the sound files compressed as files in the MP3 format (players available free for the most common machine platforms). A sound/text alignment procedure now enable the corpus user to browse the text with a standard Web-browser with hyperlinks to the sound files. It is also possible to search the text corpus at the COLT Web-site and retrieve a search result that includes hyperlinks to the relevant sound files (stored at the user's machine or delivered as sound fragments through the Internet).
The corpus was collected in five different London school boroughs in 1993 and consists of roughly half a million words of spontaneous conversation (55 hours). The conversations were recorded (surreptitiously) by student 'recruits', equipped with a Sony Walkman, a lapel microphone and a log book. The recordings have been orthographically transcribed by trained British transcribers and part of the material has been subjected to a simplified prosodic analysis. The entire corpus has been tagged for word-classes by means of the CLAWS 6 tagset developed at Lancaster University.
More information on the corpus can be found on the COLT Web-site: www.hit.uib.no/colt
Charles Meyer (University of Massachusetts, USA)
Much of the work that has been done in descriptive corpus linguistics has not been based on very sophisticated statistical analysis. Instead, many of the generalizations that have been made are based largely on frequency differences. This can lead to disastrous results, since one has no way of knowing whether frequency differences are statistically significant or not. The failure to use statistical tests in corpus analysis is largely a consequence of the fact that most corpus linguists have been trained as linguists, not as statisticians, and as a consequence have little knowledge of the appropriate tests to use on the data they are working with. In my software demonstration, I will show that one can easily use a standard statistical package such as SPSS for Windows (version 10) without knowing a lot about statistics. I will show how to code data, import it into SPSS, do various cross tabulations, and then analyze the results with various statistical tests, such as chi square and log likelihood tests.
Paul Rayson (Lancaster University, United Kingdom)
In this software demonstration, I will introduce Wmatrix, a web-based environment which allows staff and students at Lancaster local and remote access to some of UCREL's corpus annotation and retrieval tools. The web browser provides a much simpler interface to these tools than via the UNIX command line. All processing is done on the remote web server so users gain access from any platform that provides a browser such as Netscape or Internet Explorer. Tools available in Wmatrix include CLAWS (part-of-speech tagger), SEMTAG (word-sense tagger) and LEMMINGS (a lemmatiser). Wmatrix also provides production of frequency lists, statistical comparison of those lists, and KWIC concordances.
Wmatrix was built during REVERE (Rayson et al. 2000), a UK funded project investigating the extraction of information from software engineering documents. One of the aims of the project was to investigate the use of NLP tools to aid software engineers in their understanding of a software system. The information on the software system is contained in existing documentation or transcripts and reports from ethnographic studies of the system being used. We built a web-based information extraction environment by locating various UCREL NLP tools on a web server and by providing the Wmatrix interface to those tools. The output of the tools can be presented in a web browser from different viewpoints depending on the role taken by the user of the system, but this demonstration will be from the corpus linguist viewpoint. This presents the traditional model of submitting raw data to Wmatrix, passing it through the corpus annotation tools and then using concordances to view the results.
A user of Wmatrix begins by uploading their corpus to the web server via a web browser such an Netscape Navigator or Microsoft Internet Explorer. The first corpus annotation tool applied to the text is the hybrid part-of-speech tagger, CLAWS (Garside and Smith 1997) which assigns a part-of-speech tag to every word in running text with about 97% accuracy. A second layer of annotation is applied by SEMTAG, a semantic tagger (Rayson and Wilson 1996). This tool assigns a semantic field tag to every word in the text with about 92% accuracy. The resulting annotated files are presented to the user in a workarea and Wmatrix prepares word, POS and semantic tag frequency lists. These can be downloaded but can also be browsed using the web browser application. The user can select a word or tag from the lists and see a standard key word in context concordance for that item. This is prepared on the fly from the corpus on the web server.
Users are guided towards interesting words or tags to investigate further by comparing frequency lists from their corpora to standard textual norms provided by frequency lists produced from the British National Corpus for example.
Each user of Wmatrix has their own set of workareas containing corpora that they have processed. Wmatrix is designed to cope with corpora up to several million words in size, but retrieval would be less interactive with larger corpora. A web based interface for the Stuttgart Corpus WorkBench is available. The Corpus WorkBench (Christ 1994) pre-indexes the text and is consequently much faster at providing concordances for large corpora. I am currently working on integrating this into Wmatrix so that texts can be automatically indexed for CQP queries.
The REVERE project (REVerse Engineering of Requirements) is supported under the EPSRC Systems Engineering for Business Process Change (SEBPC) programme, project number GR/MO4846. Further details can be found at:
The web interface to IMS CWB was provided by Tomaz Erjavec as used in the Slovene corpus tool at: http://nl2.ijs.si/corpus/
Garside, R. and Smith, N. (1997) A Hybrid Grammatical Tagger: CLAWS4. In Garside, R., Leech, G. and McEnery, A. (eds) Corpus Annotation: Linguistic Information from Computer Text Corpora. Longman, London, pp. 102 - 121.
Rayson, P. and Wilson, A. (1996) The ACAMRIT semantic tagging system: progress report. In Evett, L. J. and Rose, T. G. (eds) Language Engineering for Document Analysis and Recognition, LEDAR, AISB96 Workshop proceedings, pp 13-20. Brighton, England.
Rayson, P., Garside, R. and Sawyer, P. (2000) Assisting requirements engineering with semantic document analysis. In Proceedings of Content-based multimedia information access RIAO 2000 (Recherche d'Informations Assistée par Ordinateur, Computer-Assisted Information Retrieval) International Conference, Collège de France, Paris, France, April 12-14, 2000. C.I.D., Paris, pp. 1363-1371.
Sean Wallis (University College London, United Kingdom)
This paper will demonstrate the new version of the ICE Corpus Utility Program, the corpus exploration program developed in conjunction with the parsed ICE-GB corpus. ICECUP has been extended in a number of directions. Fuzzy Tree Fragment queries1 support embedded logical expressions and text wild cards. The corpus map overview is supplemented by a lexicon. In addition there are major improvements in browsing results of queries.
ICECUP was initially developed as a multi-purpose ‘exploration’ platform, supporting corpus annotation. However, we have a particular priority. Linguists require a research tool that permits them to frame and investigate research questions based on the annotation in the corpus. Given, say, ICE-GB and the grammar, can we test hypotheses?
The methodology must be cyclic. Parsed corpora both permit more selective definitions of queries and tie these queries to a particular analysis scheme. Research is both limited by a framework and elucidative of the framework. There is a creative tension between corpus and theory. Results are interpreted through examples in the corpus, so the tool must support their identification and browsing. Conversely, browsing the corpus raises questions in the mind of the researcher, including criticism of the framework itself.
A parsed corpus permits us to create a diverse range of structured and extensible queries. In performing research, queries are used to specify the object of research (e.g. a particular type of verb phrase) and to identify particular outcomes (alternative special cases of the VP). We are concerned with questions like "how many of X are subclass x?" rather than their absolute frequency. This permits experiments to investigate the factors, sociolinguistic and grammatical, that correlate with the variation between alternative versions of X.
ICECUP 3.0 does not explicitly support experiments beyond performing queries and combining results. Nonetheless, it may be used for some quite sophisticated experiments2. We will demonstrate new facilities in ICECUP 3.1 to support experimentation, including the construction of frequency tables and performing statistical tests, and discuss future directions for corpus research methods.
Martin Wynne & Oliver Mason (University of Birmingham, United Kingdom)
TRACTOR is the TELRI Research Archive of Computational Tools and Resources. It features monolingual and multilingual corpora and lexicons in a wide variety of languages, currently including Bulgarian, Croatian, Czech, Dutch, English, Estonian, French, German, Greek, Hungarian, Italian, Latvian, Lithuanian, Romanian, Russian, Serbian, Slovak, Slovenian, Swedish, Turkish, Ukrainian and Uzbek. The archive, network and user community are a key part of the TELRI agenda to build links between the research communities in Western, Central and Eastern Europe. Resources distributed through TRACTOR are available for non-commercial use only, but we hope to promote and foster commercial links between academic and industrial researchers.
TRACTOR operates a User Community (TUC) for all those involved in creating, depositing and using the resources. The website www.tractor.de is the hub for the TUC. Resources can be downloaded from the website and there is a comprehensive catalogue detailing all resources and the contact details for the resource providers.
An important recent acquisition for the archive is the special TRACTOR version of the Qwick corpus analysis program which is available to the TUC with indexed monolingual corpora in many languages.
It is intended to greatly increase the range and depth of holdings in the TRACTOR archive throughout 2001, as a key part of the work of the Centre for Corpus Linguistics in the English Department at the University of Birmingham. Particular efforts are being made to create and acquire parallel translation corpora, which are of particular interest for research purposes for many academic and industrial users of TRACTOR.
The website, Qwick and some new multilingual resources will be demonstrated. Participants at ICAME 2001 are encouraged to use the archive to distribute their resources, and to join the user community!
1 Fillmore, C.J. (1992) “Corpus linguistics” or “Computer-aided armchair linguistics”. In Svartvik, J. (ed.) Directions in Corpus Linguistics. Proceedings of Nobel Symposium 82, Stockholm, 4-8 August 1991. Mouton de Gruyter, Berlin/New York, pp. 35-60.
1 A complete list of the period's news publications is found in Nelson, C & M.Seccombe. 1987. British Newspapers and Periodicals 1641-1700. A Short-Title Catalogue of Serials Printed in England, Scotland, Ireland, and British America. New York: The Modern Language Association of America.
 All the figures in brackets show the frequency per ten thousand words of the modal.
 They include the only instance of mayn't recorded in the whole corpus.
1 I do not claim that there is free variation between those three constructions and am well aware of the fact that some syntactic and semantic constraints may apply. For most cases, however, all three constructions appear to be fully idiomatic choices.
1 Telecommunications Advancement Organization of Japan.
2 Communications Research Laboratory.
1 The TOSCA-ICLE Tagger Lemmatizer was developed at the University of Nijmegen (Aarts et al. 1997).
2 The ICLE project was initiated by Prof. Sylviane Granger at the Université catholique de Louvain (Granger 1996, 1998).
1 The work presented here has grown out of a collaboration with Chris Mair and Marianne Hundt at Freiburg.
2 These latter corpora were kindly provided by the team at the Survey of English Usage.
1 Martti Mäkinen is studying English herbals from Middle English period toEarly Modern English period, and compiling a herbal corpus complementary to CEEM. The current size of the corpus is 100,000 words.
1 When speaking of learners in this text, I refer to advanced Swedish learners of English.
2 Aspect here refers to grammatical aspect.