Corpus Survey

[For an updated and expanded version of my survey, see Xiao, Richard (2008) "Well-known and influential corpora", in A. Lüdeling and M. Kyto (eds) Corpus Linguistics: An International Handbook [Volume 1]. Berlin: Mouton de Gruyter. 383-457]

As corpus building is an activity that takes times and costs money, readers may wish to use ready-made corpora to carry out their work. However, as a corpus is always designed for a particular purpose, the usefulness of a ready-made corpus must be judged with regard to the purpose to which a user intends to put it. There are thousands of corpora in the world, but most of them are created for specific research projects and are thus not publicly available. While abundant corpus resources for languages other than English are also available now, this survey focuses upon major English corpora, which are grouped in terms of their primary uses so that readers will find it easier to choose corpus resources suitable for their particular research questions. Note, however, that overlaps are inevitable in our classification. It is used in this survey simply to give a better account of the primary uses of the relevant corpora.

National corpora are normally general reference corpora which are supposed to represent the national language of a country. They are balanced with regard to genres and domains that typically represent the language under consideration. While an ideal national corpus should cover proportionally both written and spoken language, most existing national corpora and those under construction consist only of written data, as spoken data is much more difficult and expensive to capture than written data. This section introduces a number of major national corpora.

The first and best-known national corpus is perhaps the British National Corpus (BNC), which is designed to represent as wide a range of modern British English as possible so as to “make it possible to say something about language in general” (Burnard 2002, 56). The BNC comprises approximately 100 million words of written texts (90%) and transcripts of speech (10%) in modern British English. Written texts were selected using three criteria: “domain”, “time” and “medium”. Domain refers to the content type (i.e. subject field) of the text; time refers to the period of text production, while medium refers to the type of text publication such as books, periodicals or unpublished manuscripts. Table 1 summarizes the distribution of these criteria (see Aston/Burnard 1998, 29-30).

The spoken data in the BNC was collected on the basis of two criteria: “demographic” and “context-governed”. The demographic component is composed of informal encounters recorded by 124 volunteer respondents selected by age group, sex, social class and geographical region, while the context-governed component consists of more formal encounters such as meetings, lectures and radio broadcasts recorded in four broad context categories. The two components of spoken data complement each other, as many types of spoken text would not have been covered if demographic sampling techniques alone were used in data collection. Table 2 summarizes the composition of the spoken BNC. Note that in the table, the first two columns apply to both demographic and context-governed components while the third column refers to the latter component alone.

In addition to part-of-speech (POS) information, the BNC is annotated with rich metadata (i.e. contextual information) encoded according to the TEI guidelines, using ISO standard 8879. Because of its generality, as well as the use of internationally agreed standards for its encoding, the BNC corpus is a useful resource for a very wide variety of research purposes, in fields as distinct as lexicography, artificial intelligence, speech recognition and synthesis, literary studies and, of course, linguistics. There are a number of ways one can access the BNC corpus. It can be accessed online remotely using the BNC Online service or the BNCWeb interface. Alternatively, if a local copy of the corpus is available, the BNC can be explored using corpus exploration tools such as WordSmith (Scott 1999).

The current version of the full release of the BNC is BNC-2, the World Edition. This version has removed a small number of texts (less than 50) which restrict the worldwide distribution of the corpus. The BNC World has also corrected errors relating to mislabeled texts and indeterminate part-of-speech codes in the first version, and has included a classification system of genre labels developed by Lee (2001) at Lancaster. The World Edition is still marked up in TEI-compliant SGML, but an XML version of the corpus will be released shortly. As a prelude to this full release of the XML version, a four-million-word subset of the BNC – BNC Baby – was released in October 2004 together with the XML-aware corpus tool Xaira. BNC Baby was originally developed as a manageable subcorpus from the BNC for use in the language classroom, consisting of comparable samples for four kinds of English – unscripted conversation, newspapers, academic prose and written fiction (Berglund/Burnard/Wynne 2004).

The BNC model for achieving corpus balance and representativeness has been followed by a number of national corpus projects including, for example, the American National Corpus, the Polish National Corpus and the Russian Reference Corpus.

The American National Corpus (ANC) project was initiated in 1998 with the aim of building a corpus comparable to the BNC. While the ANC follows the general design of the BNC, there are differences with regarding to its sampling period and text categories. The ANC only samples language data produced from 1990 onwards whereas the sampling period for the BNC is 1960-1993. This time frame has enabled the ANC to cover text categories which have developed recently and thus were not included in the BNC, e.g. emails, web pages and chat room talks, as shown in Table 3. In addition to the BNC-like core, the ANC will also include specialized “satellite” corpora (cf. Reppen/Ide 2004, 106-107).

The ANC corpus is encoded in XML, following the guidelines of the XML version of the Corpus Encoding Standard. The standalone annotation, i.e. with the primary data and annotations kept in separate documents but linked with pointers, has enabled the corpus to be POS tagged using different tagsets (e.g. Biber’s (1988) tags, the CLAWS C5/C7 tagsets (Garside/Leech/Sampson 1987) and the Penn tags (Marcus/Santorini/Marcinkiewicz 1993) to suit the needs of different users.

The full release of the ANC is expected to be available in late 2005. At present the first release of the corpus, which contains 11.5 million words of written and spoken data (8.3 million words for writing and 3.2 million words for speech, but not balanced for genre), is now available from the Linguistic Data Consortium (LDC).

The Polish National Corpus (PNC) is under construction on the PELCRA (Polish and English Language Corpora for Research and Application) project, which is undertaken jointly by the Universities of Lodz and Lancaster. The project aims to develop a large, fully annotated reference corpus of native Polish, “mirroring the BNC in terms of genres and its coverage of written and spoken language” (Lewandowska-Tomaszczyk 2003, 106). A total of 130 million words of running texts have been collected, and part of the data (30 million words) has been compiled into a balanced corpus, which covers genres, and styles comparable in proportions to those included the BNC. The PNC is TEI-compliant and is annotated for part-of-speech. Presently, a balanced PNC sampler, which contains 10 millions of both written and spoken data reflecting proportionally the text categories in the BNC, can be ordered from the PELCRA project site.

The Czech National Corpus (CNC) consists of two sections: synchronous and diachronic. Each section is designed to include written, spoken and dialectal components. As some of the components are currently hardly more than blueprints for future work (see Kučera 2002, 254), we will only introduce the written and spoken components in the synchronous section.

The written component of the synchronous section, which contains 100 million words, was completed in 2000 and thus named SYN2000. SYN2000 includes both imaginative (15%) and informative (85%) texts, each being divided into a number of text categories, as shown in Table 4 (see Kučera 2002, 247-248). The technical and specialized texts in the corpus proportionally cover nine domains: lifestyle (5.55%), technology (4.61%), social sciences (3.67%), arts (3.48%), natural sciences (3.37%), economics/management (2.27%), law/security (0.82%), belief/religion (0.74%) and administrative texts (0.49%).

The spoken component of the synchronous section, the so-called Prague Spoken Corpus (PMK), contains 800,000 words of transcription of authentic spoken language sampled in a balanced way according to four sociolinguistic criteria: speaker sex, age, educational level and discourse type, as shown in Table 5. The data contained the Prague Spoken corpus consists exclusively of impromptu spoken language (roughly equivalent to the demographically sampled component in the BNC). Texts representing various blends of written and spoken language such as lectures, political speeches and play scripts are included in a special section in the written corpus (cf. Kučera 2002, 248, 253).

Both SYN2000 and the Prague Spoken Corpus are marked up in TEI-compliant SGML and tagged to show part-of-speech categories. SYN2000 is licensed free of charge for non-commercial use. A scaled-down version of SYN2000, PUBLIC, which contains 20 million words with the same genre distribution, is accessible online at the corpus website. The tagged version of the Prague Spoken Corpus will also be made publicly available in the near future.

The Hungarian National Corpus (HNC) is a balanced reference corpus of present-day Hungarian. The corpus contains 153.7 million words of texts produced from the mid-1990s onwards, which are divided into five subcorpora, each representing a written text type: media (52.7%), literature (9.43%), scientific texts (13.34%), official documents (12.95%) and informal texts (e.g. electronic forum discussion, 11.58%). The size of the literary subcorpus is expected to increase from the current 14.5 million words to approximately 40 million words (see Váradi 2002, 386). The HNC is encoded in SGML in compliance with Corpus Encoding Standard (CES) and annotated for part-of-speech. The corpus can be accessed free of charge after registration via the online query system at the corpus site.

The Russian Reference Corpus (BOKR) is designed as a Russian match for the BNC. The corpus contains 100 million words of modern Russian, following the general sampling frame of the BNC, as shown in Table 6 (see Sharoff 2004).

The BOKR corpus is encoded in TEI-compliant SGML and annotated for part-of-speech. As Russian is a highly inflective language, the technique used in annotating English corpora with complex POS tags is impractical for Russian, because that would entail thousands of tags which would make corpus exploration ineffective, if not impossible at all. Hence in the Russian Reference corpus, each word is annotated with a bundle of lexical and syntactic features such as part-of-speech, aspect, transitivity, voice, gender, number and tense. Separate features from a feature bundle associated with each word can be selected in a window in the query interface. The corpus is under construction and its final release is expected by the end of 2004 (cf. Sharoff 2004).

The CORIS (Corpus di Italiano Scritto) corpus is a general reference corpus of present-day Italian. It contains 100 million words of written Italian sampled from five text categories, which constitute five subcorpora, as shown in Table 7.

Unlike most national corpora that are sample corpora, the CORIS corpus follows a dynamic corpus model, which will be updated every two years by means of a built-in monitor corpus (Rossini Favretti/Tamburini/de Santis 2004). The current version of the corpus can be accessed online free of charge via a web-based query system at the corpus website.

The Hellenic National Corpus is a 32-million-word corpus of written Modern Greek sampled from several publication media covering various genres (articles, essays, literary works, reports, biographies etc.) and domains (economy, medicine, leisure, art, human sciences etc.) published from 1976 onwards. Of the four types of medium, books account for 15.75% of the total texts, newspapers 69.01%, periodicals 6.97% and miscellaneous (correspondence, electronic texts, ephemera, and hand-written/typed material) 8.27%. The text classification with regard to medium, genre and domain follows the PAROLE standards. This taxonomy information, together with the bibliographic information, is encoded in TEI-compliant SGML (cf. Hatzigeorgiu/Gavrilidou/Piperidis et al 2000, 1737). The corpus can be accessed online at the corpus site, where users can make queries concerning the lexicon, morphology, syntax and usage of Modern Greek (e.g. words, lemmas, part-of-speech categories or combinations of the three).

The German National Corpus is a product of the DWDS (Digital Dictionary of the 20th Century German Language) project. The corpus is divided into two parts, a 100-million-word balanced core and a much larger opportunistic subcorpus. This section introduces the core corpus, which is roughly comparable to the British National Corpus, covering the whole 20^th century (1900-2000). Table 8 shows the text categories covered in the corpus.

The metadata such as genre information is encoded in XML. Linguistic annotation consists basically of lemmatization, part-of-speech and semantic annotation on the word level, as well as prepositional phrase and noun phrase recognition on the phrase level (Cavar/Geyken/Neumann 2000). The core corpus is available for online search at the corpus site after free-of-charge registration.

The Slovak National Corpus is presently under construction. The project aims to create a 200-million-word corpus of the Slovak language. The first phase of the project has produced a corpus containing 30 million words of written texts published between 1990 and 2003, which will be expanded to other periods of the contemporary language (1955 – 2005) to the target size at the second phase of the project (2003-2006). The final corpus will also include diachronic and dialectological texts.

At present the 30-million-word part of the corpus has been annotated with lemmatization, morphological and source (bibliographical and style-genre) information. Users can access the corpus using a simple online query system at the corpus website. More complex searches require the “corpus manager”, which supports regular expressions and can be downloaded at the same site.

We have so far introduced national corpora for European languages. The next two sections will introduce two national corpora of Asian languages, namely Chinese and Korean.

The Modern Chinese Language Corpus (MCLC) is China’s national corpus built under the auspices of the National Language Committee of China. The corpus initially contained texts of 700 million Chinese characters sampled systematically from texts of 1.4 billion characters produced during 1919-1992 (divided into five sampling periods), with the majority of texts produced after 1977. 1919 is generally considered as the beginning of modern Chinese. Fresh data has been filtered in proportionally at the rate of 3.5 million annually since completion so that the corpus currently contains 85 million characters. The corpus covers four text categories, which include more than 40 subcategories, as shown in Table 9. Most samples are approximately 2,000 characters in length, with the exception of samples taken from books, which may contain up to 10,000 characters. The digitalized, texts were proofread three times so that errors are less than 0.02% (see Wang 2001, 283).

A scaled down version of the corpus, the core, which contains 20 million characters proportionally sampled from the larger corpus, is tokenized and tagged with part-of-speech categories. The MCLC license can be purchased from the National Language Committee of China.

The 21st Century Sejong project was launched in 1998 as a ten-year development project to build various kinds of language resources including Korean corpora and Korean electronic dictionaries. One of the goals of the project is to construct a balanced national corpus (300 million words and phrases from modern Korean, spoken materials, North Korean language, words of foreign origin, etc.), comparable to the BNC. By 2003 a raw corpus of modern Korean was compiled, containing 57 million words with 75 million more words already existing electronic texts and being processed and standardized. The corpus also includes around 3 million words of spoken data.

The markup scheme used in the Sejong Corpus is TEI-compliant. As of 2003, 10 million words have been morphologically annotated, 5.5 million words sense tagged, and 150,000 words treebanked (see Kang/Kim 2004, 1747). The corpus is accessible over the Internet after registration at the corpus site.

In addition to those introduced above, there are a number of nation-level corpora which are either already available or are under construction. They include, for example, the FRANTEXT Database, the Croatian National Corpus (30 million words), Korpus 2000 for Danish (28 million words), the National Corpus of Irish (15 million words). A number of corpora representing other national languages are also under construction, including, for example, Norwegian (Choukri 2003), Dutch (Wittenburg/Brugman/Broeder 2000), Maltese (Dalli 2001), Basque (Aduriz/Aldezabal/Alegria et al 2003), Kurdish (Gautier 1998), Nepali (Glover 1998), Tamil (Malten 1998) and Indonesia (Riza 1999).

While most of the national corpora introduced in section 2 follow a static sample corpus model, there are also corpora which are constantly updated to track rapid language change, such as the development and the life cycle of neologisms. Corpora of this type are referred to as monitor corpora.

The best-known monitor corpus is the Bank of English (BoE), which was initiated in 1991 on the COBUILD (Collins Birmingham University International Language Database) project. The corpus was designed to represent standard English as it was relevant to the needs of learners, teachers and other users, while also being of use to researchers in present-day English language. Written texts (75%) come from newspapers, magazines, fiction and non-fiction books, brochures, reports, and websites while spoken data (25%) consists of transcripts of television and radio broadcasts, meetings, interviews, discussions, and conversations. The majority of the material in the corpus represents British English (70%) while American English and other varieties account for 20% and 10% respectively. Presently the BoE contains 524 million words of written and spoken English. The corpus keeps growing with the constant addition of new material.

The BoE corpus is particularly useful for lexical and lexicographic studies, for example, tracking new words, new uses or meanings of old words, and words falling out of use. A 56 million word sampler of the corpus can be accessed online free of charge at the corpus website. Access to larger corpora is granted by special arrangement.

Another corpus of the monitor type is the Global English Monitor Corpus, which was started in late 2001 as an electronic archive of the world’s leading newspapers in English. The corpus aims at monitoring language use and semantic change in English as reflected in newspapers so as to allow for research into whether the English language discourses in Britain, the United States, Australia, Pakistan and South Africa have changed in the same way or differently. As the Global English Monitor Corpus will monitor as accurately as conceivable all relevant changes of attitudes and beliefs, it will prove a useful tool not only for lexicographers, historical linguists and semanticists, but also for those interested in social and political studies all over the world. With its first results being available at the end of 2003, the corpus is expected to reach billions of words within a few years.

The first modern corpus of English, the Brown University Standard Corpus of Present-day American English (i.e. the Brown corpus, see Kucěra/Francis 1967), was built in the early 1960s for written American English. The population from which samples for this pioneering corpus were drawn was written English text published in the United States in 1961 while its sampling frame was a list of the collection of books and periodicals in the Brown University Library and the Providence Athenaeum. The target population was first grouped into 15 text categories, from which 500 samples of approximately 2,000 words were then drawn proportionally from each text category, totaling roughly one million words.

The Brown corpus was constructed with comparative studies in mind, in the hope of setting the standard for the preparation and presentation of further bodies of data in English or in other languages. This expectation has now been realized. Since its completion, the Brown corpus model has been followed in the construction of a number of corpora for synchronic and diachronic studies as well as for cross-linguistic contrast. Table 10 shows a brief comparison of these corpora.

As can be seen, these corpora are roughly comparable but have sampled different languages or language varieties. Their sampling periods are either similar for the purposes of synchronic comparison or distanced by about three decades for the purposes of diachronic comparison. For example, the Brown and LOB (the Lancaster-Oslo-Bergen corpus of British English, see Johansson/Leech/Goodluck 1978) can be used to compare American and British English as used in the early 1960s. The updated versions of the two corpora, Frown (see Hundt/Sand/Skandera 1999) and FLOB (see Hundt/Sand/Siemund 1998) can be used to compare the two major varieties of English as used in the early 1990s. Other corpora of the similar sampling period, such as ACE (the Australian Corpus of English, also known as the Macquarie corpus), WWC (the Wellington Corpus of Written New Zealand English) and Kolhapur (the Kolhapur Corpus of Indian English), together with FLOB and Frown, allow for comparison of “world Englishes”. For diachronic studies, the Brown vs. Frown on the one hand, and the Pre-LOB, LOB and FLOB corpora on the other hand, provide a reliable basis for tracking recent language change over 30-year periods. The LCMC corpus (the Lancaster Corpus of Mandarin Chinese, see McEnery/Xiao/Mo 2003), when used in combination with FLOB/Frown corpora, provides a valuable resource for contrastive studies between Chinese and two major varieties of English.

In comparing these corpora synchronically, caution must be exercised to ensure that the sampling periods are similar. For example, comparing the Brown corpus with FLOB would involve not only language varieties but also language change. Also, as the Brown model may have been modified slightly in some of these corpora, account must be taken of such variation in comparing these corpora across text category by normalizing the raw frequencies to a common basis. Table 11 compares the text categories and number of samples for each category in these corpora.

It can be seen from the table that the two American English corpora (Brown and Frown) have the same numbers of samples for each of the 15 text categories while the British English corpora share the same proportions. The two groups differ in the numbers of samples for categories E, F, and G. The WWC and LCMC corpora follow the model of FLOB. There are important differences between the Kolhapur corpus and others in both sampling periods and the proportions of text categories. The ACE corpus covers 17 text categories instead of 15. All of these differences should be taken into account when comparing these corpora.

With the exceptions of the Pre-LOB corpus, which is under construction, and LCMC, which is distributed by the European Language Resources Association (ELRA), all of the corpora of the Brown family are available from the International Computer Archive of Modern and Medieval English (ICAME).

The corpora of the Brown family are balanced corpora representing a static snapshot of a language or language variety in a certain period. While they can be used for synchronic and diachronic studies, more appropriate resources for these kinds of research are synchronic and diachronic corpora, which will be introduced in the following two sections.

While the corpora of the Brown family are generally good for comparing language varieties such as world Englishes, the results from such a comparison must be interpreted with caution when the corpora under examination were built for different periods or the Brown model has been modified. A more reliable basis for comparing language varieties is a synchronic corpus.

A typical corpus of this type is the International Corpus of English (ICE), which is specifically designed for the synchronic study of world Englishes. The ICE corpus consists of a collection of twenty corpora of one million words each, each composed of written and spoken English produced during 1990-1994 in countries or regions in which English is a first or official language (e.g. Australia, Canada, East Africa, Hong Kong as well as Great Britain and the USA). As the primary aim of ICE is to facilitate comparative studies of English worldwide, each component follows a common corpus design as well as a common scheme for grammatical annotation to ensure direct comparability among the component corpora. All ICE corpora contain 500 texts of approximately 2,000 words each, sampled from a wide range of spoken (60%) and written (40%) genres, as shown in Table 12 (see Nelson 1996, 29-30).

Spoken (300)	Dialogues (180)	Private (100)	Conversations (90) Phone calls (10)
	Dialogues (180)	Public (80)	Class lessons (20) Broadcast discussions (20) Broadcast interviews (10) Parliamentary debates (10) Cross-examinations (10) Business transactions (10)
	Monologues (120)	Unscripted (70)	Commentaries (20) Unscripted speeches (30) Demonstrations (10) Legal presentations (10)
	Monologues (120)	Scripted (50)	Broadcast news (20) Broadcast talks (20) Non-broadcast talks (10)
Written (200)	Non-printed (50)	Student writing (20)	Student essays (10) Exam scripts (10)
	Non-printed (50)	Letters (30)	Social letters (15) Business letters (15)
	Printed (150)	Academic (40)	Humanities (10) Social sciences (10) Natural sciences (10) Technology (10)
		Popular (40)	Humanities (10) Social sciences (10) Natural sciences (10) Technology (10)
		Reportage (20)	Press reports (20)
		Instructional (20)	Administrative writing (10) Skills/hobbies (10)
		Persuasive (10)	Editorials (10)
		Creative (20)	Novels (20)

The ICE corpora are marked up and annotated at various levels. In written texts, features of the original layout are marked, including sentence and paragraph boundaries, headings, deletions, and typographic features while spoken texts are transcribed orthographically, and are marked for pauses, overlapping strings, discourse phenomena such as false starts and hesitations, and speaker turns. The bibliographic markup, which gives a complete description (e.g. text category, date, and publisher) of each text, is stored in the corpus header of each file. Different levels of annotation are undertaken for the ICE corpora. Some of them are POS tagged and parsed (e.g. the British component ICE-GB) while others are currently available as unannotated lexical corpora (e.g. the components for India, Singapore and Philippines and New Zealand). The available components of ICE can be ordered from the corpus website.

The Longman/Lancaster Corpus consists of about 30 million words of published English. British data takes up 50% and American data 40% while the other 10% represents other varieties such as Australian, African and Irish English. One half of the samples were selected randomly (“microcosmic texts”) and the other half selected by a panel of experts (“selective texts”). Most texts in the corpus are about 40,000 words long but no whole texts are used.

Both imaginative and informative text categories are included. Imaginative texts come from well-known literary works and works randomly sampled from books in print; informative texts come from the natural and social sciences, world affairs, commerce and finance, the arts, leisure, and so on. Imaginative texts are mainly works of fiction in book form while informative texts comprise books, newspapers and journals, unpublished and ephemera. Four external criteria have been used in text selection (see Holmes-Higgin/Abidi/Ahmad 1994): “region” (language varieties), “time” (1900s-1980s), “medium” (books 80%, periodicals 13.3% and ephemera 6.7%), and “level” (literary, middle and popular for imaginative texts, and technical, lay and popular for informative texts). As part of the Longman Corpus Network, the Longman/Lancaster Corpus is not available for public access.

The Longman Written American Corpus currently contains over 100 million words of running texts taken from newspapers, journals, magazines, best-selling novels, technical and scientific writing, and coffee-table books. The design of the Longman Written American Corpus is based on the general design principles of the Longman/Lancaster Corpus and the written section of the BNC. The corpus is dynamically refined and keeps growing with the constant addition of new materials. Like the other components of the Longman Corpus Network, this corpus does not appear to allow public access.

The CREA (Corpus de Referencia del Español Actual) is a corpus of standard varieties of Spanish. The corpus currently contains 133 million words sampled from a wide range of written (90%) and spoken (10%) text categories produced in all Spanish speaking countries between 1975-1999 (divided into 5-year periods). The texts in the corpus are distributed evenly between Spain and America. The domains covered in the corpus include Science and technology, social sciences, religion and thought, politics and economics, arts, leisure and ordinary Life, health, and fiction.

The CREA was designed as a monitor corpus which is continually updated so that it always represents the last twenty-five years of the history of Spanish. New data is added proportionally to maintain the corpus balance and to ensure that the various trends in current Spanish are represented. Texts for 2000-2004 are currently being incorporated (Sánchez 2002).

The CREA corpus is marked in SGML. Bibliographic and taxonomic information is encoded in the corpus header of each file. For written texts, both structural (paragraph and page number) and intratextual (notes, formulas, tables, quotations, foreign words etc.) marks are encoded. For spoken texts, the markup scheme indicates structural (speech turns) and non-structural (overlapping, tottering, anacoluthon, etc.) marks (cf. Guerra 1998).

The modular structure of the CREA corpus allows for flexible searches using geographical, generic, temporal, and thematic criteria. The corpus is accessible on the Internet.

The LIVAC (Linguistic Variation in Chinese Speech Communities) project started in 1993 with the aim of building a synchronous corpus for studying varieties of Mandarin Chinese. For this purpose, data has been collected regularly and simultaneously, once every four days since July 1995, from representative Mandarin Chinese newspapers and the electronic media of six Chinese speaking communities: Hong Kong, Taiwan, Beijing, Shanghai, Macau and Singapore. The contents of these texts typically include the editorial, and all the articles on the front page, international and local news pages, as well as features and reviews. The corpus is planned to cover a 10-year period between July 1995 and June 2005, capturing salient pre- and post-millennium evolving cultural and social fabrics of the diverse Chinese speech communities (Tsou/Tsoi/Lai et al 2000). The collection of materials from these diverse communities is synchronized with uniform calendar reference points so that all of the components are comparable. As of the end of 2003, the LIVAC corpus contained over 140 million Chinese characters, with 640,000 words in its dictionary. The corpus is expected to grow until the end of June 2005.

All of the corpus texts in LIVAC are segmented automatically and checked by hand. In addition the corpus, a lexical database is derived from the segmented texts, which includes, apart from ordinary words, those expressing new concepts or undergoing sense shifts, as well as region specific words from the six communities. The database is thus a rich resource for research into linguistics, sociolinguistics, and Chinese language and society.

As LIVAC captures the social, cultural, and linguistic developments of the six Chinese speaking communities within a decade, it allows for a wide range of comparative studies on linguistic variation in Mandarin Chinese. The corpus also provides an important resource for tracking lexical development such as the evolution of new concepts and their expressions in present-day Chinese. A sample of the corpus (data covering the period from July 1995 to June 1996) can be accessed using the online query system at the corpus site, which shows KWIC concordances as well as frequency distribution across the six speech communities.

Another way to explore language variation is from a diachronic perspective using diachronic corpora. A diachronic (or historical) corpus contains texts from the same language gathered from different time periods. Typically that period is far more extensive than that covered by Brown/Frown and LOB/FLOB or a monitor corpus such as the Bank of English. Diachronic corpora are used to track changes in language evolution. This section introduces a number of corpora of this kind.

Perhaps the best-known historical corpus is the diachronic part of the Helsinki Corpus of English Texts (i.e. the Helsinki corpus), which consists of approximately 1.5 million words of English in the form of 400 text samples, dating from the 8^th to 18^th centuries. The corpus is divided into three periods (Old, Middle, and Early Modern English) and eleven subperiods, as shown in Table 13 (cf. Kytö 1996).

In addition to the basic selection of texts as indicated in the table, there is a supplementary part in the Helsinki corpus, which focuses on regional varieties. This part consists of 834,000 words of Older Scots and 300,000 words of Old American English. While the primary selectional criteria are the dates of texts, the Helsinki corpus has sought to reflect socio-historical variation (e.g. author sex, age and social rank) and a wide range of text types (e.g. law, handbooks, science, trials, sermons, diaries, documents, plays, private and official correspondence, etc.) for each specific period. The textual markup scheme includes more than thirty genre labels, which indicate, whenever available, parameter values for the dialect and the level of formality of the text, the relationship between the writer and the receiver as well as the author’s age, sex, and social rank (Rissanen 2000).

As the Helsinki corpus not only sampled different periods covering one millennium, and it also encoded genre and sociolinguistic information, this corpus allows for researchers to go beyond simply dating and reporting language change by combining diachronic, sociolinguistic and genre studies. The Helsinki corpus can be ordered from ICAME or the Oxford Text Archive (OTA).

ARCHER, an acronym for “A Representative Corpus of Historical English Registers”, contains 1.7 million words of data in the form of 1,037 texts sampled from seven 50-year historical periods covering Early Modern English (1650-1990). The corpus is designed as a balanced representation of seven written (journal-diaries, letters, fiction, news, and science, etc.) and three speech-based (fictional conversation, drama and sermons-homilies) genres in British (two thirds of the corpus) and American (one third, data available only for the periods 1750-1799, 1850-1899, 1950-1990) English. Each 50-year subcorpus includes 20,000-30,000 words per register, typically containing ten texts of approximately 2,000-3,000 words each (cf. Biber/Finegan/Atkinson 1994). ARCHER is tagged for grammatical/functional categories. It allows for a wide variety of investigations on recent linguistic change and change in discourse and genre conventions. The corpus is presently being expanded with more American texts to make the American and British data comparable (see ARCHER 2). The expanded version will also enable a systematic comparison of the two varieties of English diachronically. However, because of the copyright problem, ARCHER is not publicly available at the moment. Readers interested in using this corpus can contact Douglas Biber.

In addition to the Helsinki and ARCHER corpora, which cover many centuries, there are a number of well-known historical corpora focusing a particular period or a specific domain or genre, which will be introduced in the following sections.

The Lampeter Corpus of Early Modern English Tracts is a balanced corpus covering one century between 1640 and 1740, which is divided into ten decades. Each decade consists of data sampled from six domains (religion, politics, economics/trade, science, law and miscellaneous). Two complete texts, ranging from 3,000 to 20,000 words, are included for each domain within each decade, totaling approximately 1.1 million words (Schmied 1994).

The Lampeter corpus is encoded in TEI-compliant SGML. The TEI headers provides the framework for historical, sociolinguistic and stylistic investigations, including information regarding authors (name, age, sex, place of residence, education, social status, political affiliation), printers/publishers, place and date of print, publication format, text characteristics and bibliographical sources. As the corpus includes whole texts rather than smaller samples, the corpus is also useful for study of textual organization in Early Modern English. The Lampeter corpus can be ordered from ICAME or OTA.

The Dictionary of Old English Corpus in Electronic Form (DOEC, the 2000 release) contains 3,037 texts of Old English, totaling over three million words, in addition to two million words of Latin. The texts in the corpus are practically all extant Old English writings. The DOEC corpus includes at least one copy of each surviving text in Old English while in cases where it is significant because of dialect or date, more than one copy is included. These texts cover six text categories: poetry, prose, interlinear glosses, glossaries, runic inscriptions, and inscriptions in the Latin alphabet. In the prose category in particular, a wide range of text types are covered which include, for example, saints’ lives, sermons, biblical translations, penitential writings, laws, charters and wills, records (of manumissions, land grants, land sales, land surveys), chronicles, a set of tables for computing the moveable feasts of the Church calendar and for astrological calculations, medical texts, prognostics (the Anglo-Saxon equivalent of the horoscope), charms (such as those for a toothache or for an easy labour), and even cryptograms (cf. the corpus website). The texts in the corpus are encoded in TEI-compliant SGML. The DOEC corpus can be ordered on CDs or assessed online by institutional site license at the corpus website. The web-based query system allows for searches by single words, word combinations, word proximity and bibliographic sources.

Early English Books Online (EEBO) is a joint effort launched in 1999 between the University of Michigan, Oxford University and ProQuest Information and Learning to create a full-text archive of Early English. From the first book published in English through the age of Spenser and Shakespeare, the EEBO collection now contains about 100,000 of over 125,000 titles listed in Pollard & Redgrave’s Short-Title Catalogue (1475-1640) and Wing’s Short-Title Catalogue (1641-1700) and their revised editions, as well as the Thomason Tracts (1640-1661) collection and the Early English Books Tract Supplement, covering a wide range of domains including, for example, English literature, history, philosophy, linguistics, theology, music, fine arts, education, mathematics and science. The remaining titles will be digitalized and added to the database in the near future. The corpus can be accessed online at the EEBO website.

The Corpus of Early English Correspondence (CEEC, the 1998 version) consists of 96 collections of ca. 6,000 personal letters written by 778 people (women accounting for 20%) between 1417 and 1681, totaling 2.7 million words. The corpus is accompanied by a sender database, which offers users easy access to various sociolinguistic variables, including writer age, gender, place of birth, education, occupation, social rank, domicile and the relationship with the addressee. CEEC is a balanced corpus which can be neatly divided into two parts, both covering chronologically fairly equal periods: the first from ca. 1417 to 1550 and the second from 1551 to 1680 (cf. Laitinen 2002). Table 14 shows the proportions in terms of writers’ social ranks and domiciles (see Nevalainen 2000: 40). The CEEC corpus is currently being expanded with personal letters written between 1682 and 1800 to cover the 18th-century.

As the copyright problem has prevented public access to the full release of the CEEC corpus, a CEEC sampler (CEECS) has been published by ICAME, which represents the non-copyrighted materials included CEEC. The sampler reflects the structure of the full CEEC only in some respects. The time covered is nearly the same (1418-1680), which is divided into two parts. CEECS1 (246,055 words) covers the 15th and 16th centuries while CEECS2 (204,030 words) covers the 17th century. The sampler corpus consists of 23 collections of 1,147 letters with 194 informants, totaling 450,085 words. The CEEC sampler is available from ICAME or OTA.

The Zurich English Newspaper Corpus (ZEN) is a 1.2-million-word collection of newspapers in Early English, covering 120 years (from 1671 to 1791) of British newspaper history. To achieve a representative coverage, a wide variety of newspapers were included. Up to ten issues per newspaper were selected at ten-year intervals throughout the whole period. With the exception of stock market reports, lottery figures, long lists of names and poetry, the whole newspapers were included in the corpus. The news stories are grouped into two major categories: foreign news and home news, with each news category further classified according to its own text genre definition (cf. Fries/Schneider 2000). The corpus is split into four 30-year periods in order to track potential language change, as shown in Table 15 (see Schneider 2002: 202).

The ZEN corpus is SGML-conformant. It not only allows for linguistic analysis of different types of news stories in the 17^th and 18^th centuries, it has also made it possible to compare news texts in Early English with modern newspaper language. The ZEN query system allows restricted access to the online database.

The Innsbruck Computer Archive of Machine-Readable English Texts (ICAMET) contains ca. 500 Middle English texts totaling 5.7 million words. The database comprises three parts, namely, the Prose Corpus (129 texts written during 1100-1500, accounting for two thirds of the total), the Letter Corpus (254 letters written during 1386-1688, arranged in the diachronic order), and the Prose Varia Corpus (mainly translations or normalized versions of Middle English texts). An advantage of ICAMET is that the database consists of complete texts instead of extracts, which allows literary, historical and topical analyses of various kinds, particularly studies of cultural history (Marcus 1999). Nevertheless, the copyright issue has restricted public access to many prose texts in the corpus. A sampler containing half of the prose texts and all letters is available from ICAME.

The Corpus of English Dialogues (CED) contains 1.3 million words of Early Modern English dialogue texts produced over a 200-year time span between 1560 and 1760. While the spoken language of the past is inaccessible directly to modern speakers, it is recorded in speech related texts. The CED corpus sampled from six such text categories, including trial proceedings, witness depositions, drama, handbooks in dialogue form, fictional dialogues, and language teaching books (cf. Culpeper/Kytö 1997).

The focus on dialogue will allow insight into the nature of impromptu speech and interactive two-way communication in the Early Modern English period - aspects which have received little research attention. The CED corpus is currently under construction by the Universities of Lancaster and Uppsala.

A Corpus of Late Eighteenth-Century Prose contains 30,000 words of unpublished letters transcribed from the originals dated from the period 1761-1790. The corpus is distributed in both plain text (extended ASCII) and HTML versions. The text version can be used with a concordancer while the HTML version facilitates viewing the corpus in a browser. The plain text version is marked up in the COCOA format, giving information on writer, date and page breaks, etc. The corpus is intended to complement major diachronic corpora like the Helsinki corpus, which stop in the early eighteenth century. Another aim of the corpus is “to illustrate non-literary English and English relatively uninfluenced by prescriptivist ideas, in the belief that it might help with research into change in (ordinary, spoken) language in the late Modern English period” (van Bergen/Denison 2004). The corpus is by no means uniform, nor is it balanced. Nevertheless, because of the nature of the material, it is of great use to both linguists and historians. The corpus can be ordered from the Oxford Text Archive, free of charge, for use in education and research.

A Corpus of Late Modern English Prose contains 10,000 words of informal private letters written by British writers between 1861 and 1919. All decades in this period are represented, with about 6,000 words for the decade 1880-1889, 13,000 words for 1890-1899 and 20,000 words for the other four decades each. These blocks of texts are sampled from five sources.

Stored in seven extended (8-bit) ASCII text files, the corpus is marked up following the conventions used in the Helsinki corpus, with information on writer, recipient, relationship, date, genre, and page etc. encoded in COCOA-style brackets (see Denison 1994). The corpus can be ordered at no cost from the Oxford Text Archive.

While general corpora like national corpora may contain spoken material, there are a number of well-known publicly available spoken English corpora, which will be introduced in this section.

The London-Lund Corpus (LLC), as the first electronic corpus of spontaneous language, is a corpus of spoken British English recorded from 1953-1987. The corpus derived from two projects: the Survey of English Usage (SEU) at University College London and the Survey of Spoken English (SSE) at Lund University. There are two versions of LLC, the original version consisting of 87 transcripts from SSE totaling 435,000 words, and the complete version, which has been augmented by 13 supplementary transcripts from SEU. The full LLC corpus comprises 100 texts, each of 5,000 words, totaling half a million running words. A distinction is made between dialogue (e.g. face-to-face conversations, telephone conversations, and public discussion) and monologue (both spontaneous and prepared) in the organization of the corpus (cf. Greenbaum/Svartvik 1990). This textual information is encoded together with speaker information (e.g. gender, age, occupation). The texts in the corpus are transcribed orthographically, with detailed prosodic annotation. The LLC corpus is available from ICAME.

The Lancaster/IBM Spoken English Corpus (SEC) consists of approximately 53,000 words of spoken British English, mainly taken from radio broadcasts dating between 1984 and 1991. For a corpus of this size, it is impossible to include samples of every style of spoken English. The SEC corpus has been designed to cover speech categories suitable for speech synthesis, as shown in Table 16 (see Taylor/Knowles 1988).

In the SEC corpus, efforts have been made to achieve a balance between the highly stylized texts (e.g. poetry, religious broadcast, propaganda) and dialogue, and between male and female speakers. Of the 53 speakers in the corpus, 17 are female, representing 30% of the corpus. The higher proportions of male speakers in the news and commentary categories reflect the tendency of the BBC to use mainly male speakers in these types of programmes.

SEC is available in orthographic, prosodic, grammatically tagged and treebank versions, which should prove most useful to those who research in the speech synthesis or speech recognition fields. The corpus can be ordered from ICAME.

The Machine Readable Spoken English Corpus (MARSEC) is an extension of SEC in which the original acoustic recordings were digitalized, and word-level time-alignment between the transcripts and the acoustic signals was included. Tonetic stress marks were also converted into ASCII symbols to make the corpus machine-readable. The prosodically annotated word-level alignment files are available at the MARSEC website.

The Aix-MARSEC database is a further development of MARSEC. The database consists of two major components: the digitalized recordings from MARSEC and the annotations. Annotations have so far been undertaken at nine levels such as phonemes, syllables, words, stress feet, rhythm units, minor and major turn units. Two supplementary levels, the grammatical annotation by CLAWS and a Property Grammar system developed at Aix-en-Provence, are to be integrated soon (cf. Auran/Bouzon/Hirst 2004). The database, together with tools, is available under GNU GPL licensing at the Aix-MARSEC project site.

The Bergen Corpus of London Teenage Language (COLT) is the first large English corpus focusing on the speech of teenagers. It contains half a million words (about 55 hours of recording) of orthographically transcribed spontaneous teenage talk recorded in 1993 by 31 volunteer recruits from five socially different school boroughs. The speakers in the corpus are classified into six age groups: preadolescence (0-9 years old), early adolescence (10-13), middle adolescence (14-16), late adolescence (17-19), young adults (20-29) and older adults (30+). As the name of corpus suggests, the core of the corpus represents teenagers. The early, middle and late adolescence groups account respectively for 24%, 61% and 9%, totaling 94% of the corpus. The older adult group, mostly parents, teachers, takes up 6%. As regards speaker gender, girls and boys contributed roughly the same amount of text: the male speakers about 51.8% (230,616 words) and the female speakers 48.2% (214,215 words). In terms of social class, only about 50% of the corpus material can be assigned a social group value. The material that has been classified is evenly distributed across the three social groups: high, middle, and low. While a wide range of settings are present in the COLT corpus, settings in connection with school (48%) and home (32%) are the most common. Such speaker-specific information (speaker age, gender, social class, etc.) and conversation-specific information (location and setting) is encoded in the header of each corpus text. In the body of the text, paralinguistic features and non-verbal sounds are also marked up (cf. Haslerud/Stenström 1995).

The corpus constitutes part of the British National Corpus. In addition, COLT is released in both orthographically transcribed (pure text) and tagged version (using CLAWS C7 tagset). A prosodically annotated version (a representative selection amounting to approximately 150,000 words) is also available. The corpus is for non-commercial purposes and can be accessed online by registered users or ordered form ICAME.

The Cambridge and Nottingham Corpus of Discourse in English (CANCODE) is part of the Cambridge International Corpus (CIC, see Appendix). The corpus comprises five million words of transcribed spontaneous speech recorded in Britain and Ireland between 1994 and 2001, covering a wide variety of mostly informal settings: casual conversation, people working together, people shopping, people finding out information, discussions and many other types of interaction. As CANCODE is designed as a contextually and interactively differentiated corpus, the data has been carefully collected and sociolinguistically profiled with reference to a range of different speech genres and with an emphasis on everyday communication.

A unique feature of CANCODE is that the corpus has been coded with information pertaining to the relationship between the speakers: whether they are intimates (living together), casual acquaintances, colleagues at work, or strangers. For this purpose, CANOCDE is organized along two main axes: context-type and interaction-type. Alongside the axis of context-type are, on the cline from “public” to “private”, transactional, professional, socializing and intimate. Alongside the axis of interaction-type, on the cline from “collaborative” to “non-collaborative”, information provision, collaborative idea, and collaborative work. The interactions between the two axes, together with typical settings, are shown in Table 17 (see Carter/McCarthy 2004, 67). This coding allows users to look more closely at how different levels of familiarity (formality) affect the way in which people speak to each other. The corpus is not currently available to the public.

A corpus that was built specifically for the study of English dialects is the spoken corpus of the Survey of English Dialects (SED, see Beare/Scott 1999). The Survey of English Dialects was started in 1948 by Harold Orton at the University of Leeds. The initial work comprised a questionnaire-based survey of traditional dialects based on extensive interviews of about 1,000 people from 313 locations all over rural England. During the survey, a number of recordings were made as well as the detailed interviews. The recordings, which were made during 1948-1961, consist of about 60 hours of dialogue of people aged 60 or above talking about their memories, families, work and the folklore of the countryside from a century ago. Elderly people were chosen as subjects because they were most likely to speak the traditional, “uncontaminated” dialect of their area.

The spoken corpus derived from SED consists of transcripts of 314 recordings from 289 (out of the 313) SED localities in England, totaling roughly 800,000 running words. The original recordings were transcribed, with sound files linked to transcripts. The corpus in TEI-compliant SGML and POS tagged using CLAWS.

While the spoken corpus of SED comprises data invariably produced by elderly people, as the survey was conducted nationwide, covering every county of England, it has, for the first time, made it possible to conduct a detailed study of the regional variation in English dialects on a national level. Also, as the data reflects a society which was different in many ways from today, the corpus is a valuable resource for dialectologists, historical linguistics as well as historians. The CD-ROM of the spoken corpus is published by Routledge, London.

The Intonational Variation in English (IviE) corpus was constructed for the investigation of cross-varietal and stylistic variation in British English intonation, focusing on nine urban varieties of English spoken in the British Isles, i.e. Belfast, Bradford, Cambridge, Cardiff, Dublin, Liverpool, Leeds, London, and Newcastle. The corpus comprises 36 hours of speech data in five different speaking styles: phonetically controlled sentences (statements, questions without morpho-syntactic markers, WH-questions, inversion questions, coordination structures), a read text (the fairy tale Cinderella), a retold version of the same text, a map task (“find your way around a small town”) and free conversations (on the assigned topic of smoking). The data was collected in urban secondary schools, and the speakers were 16 years old at the time when the recordings were made. A minimum of six male and six female speakers from each variety were recorded, though more speakers were included for some of the varieties, totaling 116 speakers in all (cf. Grabe/Post/Nolan 2001). The corpus is available free of charge for non-commercial use only. Orthographic and prosodic transcriptions, together with digitalized sound files can be ordered on CDs or downloaded from the corpus website.

The Longman British Spoken Corpus contains 10 million words of natural, spontaneous conversations from a representative sample of the population in terms of speaker age, gender, social group and region, and from the language of lectures, business meetings, after dinner speeches and chat shows. The design criteria are discussed in detail in Crowdy (1993). The Longman British Spoken Corpus is the first large scale attempt to collect spoken data in a systematic way. The corpus is part of the spoken section of the British National Corpus (see section 2.1).

The Longman Spoken American Corpus comprises five million words of spoken data collected from everyday conversations of more than 1,000 Americans of various age groups, levels of education, and ethnicity from over 30 US States. Equal numbers of participants were chosen from each region, and a balance was struck between the numbers of participants from rural and city areas within those regions. Recordings were made of four-hour chunks of the normal daily conversations of each participant over periods of at least four days. The participants were chosen to be representative for gender, age, ethnicity and education, as shown by the latest US demographic census statistics (Table 18, see Stern 1997). As part of the Longman Corpus Network, the Longman Spoken American Corpus is a property of the Longman publishers for in-house use only.

The Santa Barbara Corpus of Spoken American English (SBCSAE) is based on hundreds of recordings of spontaneous speech from all over the United States, representing a wide variety of people of different regional origins, ages, occupations, and ethnic and social backgrounds. It reflects the many ways that people use language in their lives: conversation, gossip, arguments, on-the-job talk, card games, city council meetings, sales pitches, classroom lectures, political speeches, bedtime stories, sermons, weddings, etc. (cf. Dubois/Chafe/Meyer et al 2000-2004).

The corpus is particularly useful for research into speech recognition as each speech file is accompanied by a transcript in which phrases are time stamped to allow them to be linked with the audio recording from which the transcription was produced. Personal names, place names, phone numbers, etc, in the transcripts have been altered to preserve the anonymity of the speakers and their acquaintances and the audio files have been filtered to make these portions of the recordings unrecognisable. The SBCSAE corpus is distributed by the LDC in five parts, the first three of which have been released to date.

The Saarbrücken Corpus of Spoken English (SCoSE) consists of three parts: stories, jokes and interviews. The first two parts comprise excerpts transcribed from audio-taped talk recorded by researchers and students at Northern Illinois University and at Saarland University. Most of the excerpts come from real conversations among family members and friends, fellow students and colleagues. The third part includes transcripts of stories recorded in interviews with senior citizens aged 80 and older in a retirement community in Indianapolis, Indiana in the summer of 2002. In all of these parts, speech turns are indicated. The hard copy of corpus (in PDF format), together with a description of transcription conventions, is available at no cost at the corpus site. The electronic copy of the corpus is downloadable and accessible online at Talkbank. Users can use the interface at the Talkbank site to generate three different versions to suit their research interests: the properly marked up text version, the XML version, and the HTML version in which each utterance is aligned with audio recording.

The Switchboard Corpus (SWB) is a corpus of is 2,438 spontaneous telephone conversations, averaging 6 minutes in length, recorded for over 542 speakers of both sexes from every major dialect of American English in the early 1990s. The transcripts total three million words (over 240 hours of recordings). Information relevant to speakers' sex, year of birth, education level and dialect region is available in the documentation accompanying the corpus. Table 19 shows the distribution of major sociolinguistic variables (see Godfrey/Holliman 1997).

As each transcript in the corpus is time-aligned at the word level, the corpus is useful for sociolinguistic studies as well for speech recognition. The corpus is distributed by the LDC. It can also be downloaded from the Switchboard website or accessed via the LDC Online.

The Wellington Corpus of Spoken New Zealand English (WSC) comprises one million words of spoken New Zealand English in the form of 551 2,000-word extracts collected between 1988 and 1994 (99% of the data from 1990–1994, the exception being eight private interviews). A very stringent criterion was adopted to ensure the integrity of the New Zealand samples included in the corpus. Data was collected only from those who lived in New Zealand before the age of 10, those who had spent less than 10 years (or half lifetime, whichever was greater), and those who made an overseas trip over one year before. The extracts are classified into 15 text categories covering a wide range of contexts in which each style of speech is found, as shown in Table 20 (cf. Holmes/Vine/Johnson 1998).

The formal speech section (12%) in the WSC corpus includes all monologue categories and “parliamentary debate” in the public dialogue category. The semi-formal section (13%) includes the three types of interview (both public and private). All of the other text categories make up the informal speech section (75%), with private conversation alone accounting for 50% of the corpus. In terms of speaker gender, women contributed 52% and men 48% of the final transcribed words, reflecting the New Zealand population balance. With regard to speaker age, data for the age group 20-24 accounts for more than 20% of the corpus, and the proportions for age groups 45-49 and 40-44 both exceed 10% while there is little data for those aged over 70. The distribution across different age groups generally mirrors the population structure in New Zealand. The corpus data also reflects the distribution of population across ethnic groups, with data collected for Pakeha accounting for 76%, and for Maori 18%. Every speech sample included in the corpus is described as fully as possible in terms of sociolinguistic variables such as the gender, age, regional origin, social class, level of education and occupation of its contributor.

The unusually high proportion of private material and the rich sociolinguistic variation make the WSC corpus a valuable resource for research into informal spoken registers as well as for sociolinguistic studies. The corpus is available from ICAME.

The Limerick corpus of Irish English (L-CIE) comprises one million words in the form of 375 transcripts of naturally occurring conversations recorded in a wide variety of speech contexts throughout Ireland (excluding Northern Ireland). Speakers range from 14 to 78 years of age and there is an equal representation of both male and female speakers. While the corpus consists mainly of casual conversation, it also has over 200,000 words of professional, transactional and pedagogic Irish English which, along with the casual conversation data, were carefully collected with reference to a range of different speech genres. The corpus has followed the design of CANCODE by organizing the corpus alongside the axes of context type and interaction type, as shown in Table 21 (cf. Farr/Murphy/O’Keeffe forthcoming).

While it is not designed to be geographically representative – it does not include data from every county in the Republic of Ireland, the L-CIE corpus has developed a careful sociolinguistic classification scheme which facilitates inter-corpus comparisons, especially with regard to linguistic choices and the relationships that hold between the speakers. The corpus website allows online access by registered users.

The Hong Kong Corpus of conversational English (HKCCE) comprises 50 hours of recordings made up of 130 separate conversations involving a total of 341 participants. The lengths of the conversations are between 1 hour 15 minutes and 2 minutes 49 seconds, averaging about 23 minutes in length. The corpus is divided into four subcorpora (conversations, academic discourses, business discourses and public discourses), amounting to approximately 500,000 words. The recordings were made in the mid-1990s of conversations between Hong Kong Chinese and non–Cantonese speakers (mostly native speakers of English). Table 22 shows the distribution of the data cross various design criteria (cf. Cheng/Warren 1999, 13-16).

The corpus has not only facilitated sociolinguistic research in English spoken in Hong Kong, it has also made it possible to compare native and non-native spoken English. In addition to the orthographic transcription, the corpus is currently being annotated prosodically to enable them to examine the communicative role of intonation. The corpus has not been publicly released.

As language may vary considerably across genre and domain, specialized corpora provide valuable resources for investigations in the relevant genres and domains. Unsurprisingly, there has recently been much interest in the creation and exploitation of specialized corpora in academic or professional settings. This section introduces a number of well-known English corpora of this kind.

The Michigan Corpus of Academic Spoken English (MICASE) contains approximately 1.8 million words in the form of 152 transcripts of nearly 200 hours of recordings of 1,571 speakers, focusing on contemporary university speech within the domain of the University of Michigan. Table 23 shows the structure of the corpus (cf. MICASE Manual).

In the MICASE corpus, speakers are divided into four age groups: 17-23, 24-30, 31-50, and 51+. In terms of academic role, they are classified into a number of categories: junior and senior undergraduates, junior and senior postgraduates, junior and senior faculty and researchers, etc. The language status can be native speaker (North American English), other native speaker (non-American English), near native speaker, and non-native speaker.

The MICASE corpus was originally marked up in TEI-compliant SGML. All of the SGML files have now been converted to the XML format in order to meet the requirements for further corpus development including a web-based search interface and the streaming web delivery of the sound recordings, synchronized with the transcripts. At present, only the orthographically transcribed version of the corpus is available, though future releases will include various kinds of annotations such as parts-of-speech, lemmas and discourse-pragmatic categories. The MICASE corpus can be searched online free of charge or ordered at a nominal fee at the corpus website.

The British Academic Spoken English (BASE) corpus, which is designed as a British counterpart to the MICASE, is under construction at the Universities of Reading and Warwick. The corpus currently comprises a collection of recordings and marked up transcripts of 160 lectures (63 from Reading and 97 from Warwick, totaling 127 recording hours) and 39 seminars (from Warwick, 32 hours). The lectures and seminars spread evenly across four subject areas, as shown in Table 24 (cf. the corpus website).