 |
British National Corpus User Reference Guide |
2. Design of BNC-baby | |
| Author: edited by Lou Burnard (revised LB) Date: (revised 19-22 Nov 2003) |
Up: Contents Previous: 1. Introduction Next: 3. Basic structure
BNC-baby is a four-part corpus constructed by a principled sampling taken from the BNC World Edition. The principles underlying selection of texts for inclusion in BNC-baby may be summarized as follows:
- equal amounts of material (approximately one million words each) have been selected from the full BNC as representative of academic writing, imaginative writing, newspaper texts, and spontaneous conversation.
- texts previously included in the BNC Sampler were excluded from selection.
To select materials meeting these criteria, the following procedure was adopted. Selection of texts was made on the basis of the information about them recorded in the original corpus, in particular the text classifications provided by David Lee.
2.1. Fiction
The original BNCworld contains 477 texts (16 million words) classed as `imaginative'. These vary considerably in date of publication, target audience and medium, as well as in author properties and other features. Compared with other texts in the corpus, these texts are also rather long. Achieving a comparable variety in the one million word subset would be difficult, if not impossible. In BNC-baby therefore, we selected according to the following carefully defined criteria. Texts for the fiction component were selected from among all the BNCworld written imaginative texts those which were published as books between 1985-1994 and classified explicitly as for an adult audience and were assigned the genre label W fict prose. From this set of 356 texts, a random sample of about one million words (25 texts) was drawn. The sample was checked to ensure no more than one title by any particular author was selected.
2.2. Newspapers
In the design of the original BNC, newspapers were not identified as a distinct category, although a very large amount of newspaper material is contained in it. On the basis of the descriptive information provided in the text headers, it is however possible to select different kinds of newspaper text. Note that each `text' is made up of several newspaper articles, often drawn from a particular subject domain, rather than complete issues of a paper. In this, BNC-baby follows the sampling methods of the original corpus compilers.
The BNC-baby selection was made to ensure a mix of national and local papers, wide coverage of topics and little duplication of dates. Approximately 60% of the newspaper data comes from five national papers, and the remaining 40% from regional newspapers. An attempt was made to include texts from different domains and genres, to maximise the spread in topic areas covered. As far as possible, the texts were selected to ensure a spread of dates of publication in order to minimise the effects of seasonal or topical variation. The size of the texts was considered and choices made to ensure a roughly equal distribution across different newspapers within the national and regional subsets respectively, as well as a spread across subject areas. Differences between the amount of data from each newspaper in the component is largely due to the considerable variation in the size of the newspaper texts in BNCworld.
The following table shows the number of words in each newspaper sampled for BNC-baby
| Newspaper | words | % |
| Daily Mirror | 124251 | 12% |
| Daily Telegraph | 128794 | 13% |
| Guardian | 129598 | 13% |
| Independent | 131205 | 13% |
| Today | 91238 | 9% |
| Belfast Telegraph | 43006 | 4% |
| East Anglian Daily Times | 43674 | 4% |
| Liverpool Daily Post and Echo | 85441 | 9% |
| Northern Echo | 68887 | 7% |
| The Alton Herald | 56316 | 6% |
| The East Anglian | 15814 | 2% |
| The Scotsman | 66709 | 7% |
| Ulster Newsletter | 16888 | 2% |
| Total component | 1001821 | 100% |
2.3. Spoken
The conversational data in the BNC-baby has been drawn only from the spoken demographic component of the BNC World edition. Each text consists of a number of conversations recorded by one individual, capturing data produced by a number of different speakers in different situations. Speakers were recruited (as described elsewhere) according to demographic principles, in order to be broadly representative of the UK population in terms of age, gender, region, and class.
Texts for which very little information about the speakers was available were excluded from selection. From the remainder, 30 texts were then randomly selected.
The following table shows the number of words spoken by participants in the spoken part of BNC-baby, broken down by sex and age group:
| category | value | words | % of categoru | % of corpus |
| Age of speaker | 0-14 | 102,350 | 11% | 10% |
| 15-24 | 73,891 | 8% | 7% | |
| 35-44 | 182,976 | 20% | 18% | |
| 25-34 | 292,083 | 32% | 29% | |
| 45-59 | 113,038 | 12% | 11% | |
| 60+ | 159,948 | 17% | 16% | |
| total | 924,286 | 100% | 91% | |
| Sex of speaker | Female | 551,077 | 59% | 55% |
| Male | 384,337 | 41% | 38% | |
| total | 935,414 | 100% | 93% | |
| Social class of speaker | AB | 243,125 | 35% | 24% |
| C1 | 232,981 | 33% | 23% | |
| C2 | 133,963 | 19% | 13% | |
| DE | 86,189 | 12% | 9% | |
| total | 696,258 | 100% | 69% |
2.4. Academic
In the design of the original BNC, academic prose is not identified as a distinct category, although a large amount of such material is contained in it. On the basis of the descriptive information provided in the text headers, it is however possible to select such texts. From the set of texts identified by David Lee as "written academic", titles were randomly selected within different subject areas to maximize variation in topic. An attempt was also made to include data originally published in periodicals as well as in books, although no targets were set for the proportions of material from each medium. Of the 501 academic writing texts in the BNCworld, 30 were selected for the BNC-baby academic component.
2.5. Design of the BNC World Edition
Sections 2 to 4 of the Users Reference Guide supplied with the BNC World Edition give a detailed overview of the design principles underlying the original construction of the British National Corpus, which principles are necessarily followed in the sampling for BNC-baby. The BNC World Users Reference Guide also includes detailed information about the actual composition of the full corpus with respect to its selection and classification.
We do not duplicate that information here: it should however be consulted for a proper understanding of the composition of the BNC-baby corpus.
Up: Contents Previous: 1. Introduction Next: 3. Basic structure
Date: (revised 19-22 Nov 2003) Author: edited by Lou Burnard (revised LB).
British National Corpus.