THE BNC The British National Corpus

1
THE BNCThe British National Corpus

• TU Chemnitz
• English Language and Linguistics
• HS Practical Corpus Linguistics
• Lecturer Martin Weisser, Ph.D.
• Presented by Kristin Flanagan
• Wintersemester 2005/2006

2
What is the BNC?

• corpus of more than 100 million words of
  contemporary spoken and written British English
• representative of British English as a whole
• covers wide range of genres from written and
  spoken English for educational, academic and
  commercial use

• was put together by Oxford University Press,
  Longman Group (UK) Ltd., W. R. Chambers, the
  British Library, the Universities of Oxford and
  Lancaster
• 1991 1995

3
How is the BNC divided?

• 4,124 texts based on 90 written, 10 spoken
  sources
• Written
• texts of written sources come from
• 75 informative prose, all post 1975
• 25 imaginative (literary works), all post 1960

• 60 taken from books
• 25 from periodicals
• 5 from public brochures
• 5 from unpublished letters, minutes, essays
• 5 written plays or speeches
• written corpus represents different levels of
  British English
• 30 more literary or technical high style
• 45 middle
• 25 informal low style

4
The spoken part of the BNC

• all spoken data dates back no earlier than 1991
• two major sources
• 1. context-dependent section
• educational and informative events
• news reports
• business events
• official and public events
• leisure events

• 2. demographic section (40)
• 2,000 hours of transcribed recordings
• made by 127 volunteers
• all their conversations from a two to seven day
  period
• from 38 different parts of the UK
• four different socio-economic groupings
• balanced coverage of female and male speakers
• ages between 15 and 60

5
Modifications made to Claws to deal with spoken
data

• additions to the lexicon
• - interjections of various sorts (ah,
  mhm, okey-dokey, etc.)
• - various taboo words
• - slang words
• - small list of words used differently spoken
  vs. written
• addition to the contractions recognized by Claws
  (dya, gotta, etc.)
• special treatment of truncated words and some
  interjections
• recognition of repetition

6
Grammatical tagging of the spoken part of the BNC

• corpus is being encoded using SGML (Standard
  Generalized Markup Language) in conformance with
  TEI (Text Encoding Initiative)
• every word is assigned part-of-speech markers
  from tagset of 61 tags, which is called C5 tagset
• tagging is carried out by Claws4 tagging system
• after Claws4 manual post-editing phase
• finally, sent back to Oxford University Computing
  Services for possible errors

7
Tagging

• utterances are indicated with pair of SGML tags
  ltugt and lt/ugt
• indication of speaker is given by who attribute
• overlapped passages indicated with SGML ltptrgt
  tags
• laughs, pauses, etc. indicated with appropriate
  SGML tag
• after processing by UCREL, text has been split
  into s-units (essentially functional sentences)
• each sentence is given identifying code number
• each word is assigned part-of-speech marker

8
Sound example

• ltcreation date'?'gt
• Origination/creation date not known
• lt/creationgt
• ltparticsgt
• ltperson age5 dialectXLO educ2
  flangEN-GBR idPS0SV nW0001
• roleother sexf socC1gt
• Age 66
• BMRB code 101
• BNC name Sidney
• Name Irene
• Occupation retired
• lt/persongt
• ltperson ageX dialectXLO educX
  flangEN-GBR idPS140 nW0002
• respPS0SV roleother sexf socUUgt
• Name Ethel
• lt/persongt
• ltperson age5 dialectXLO educX
  flangEN-GBR idPS141 nW0003
• respPS0SV roleother sexf socUUgt
• Age 67

9
Transcription of the previous sound example
(excerpt)

• ltu whoPS140gt
• lts n0068gt
• ltw AV0gtWell ltw CJSgtonce ltw PNPgtyoultw VHBgt've ltw
  VDNgtdone ltw DT0gtthat
• ltw PNPgtitltw VBZgt's ltw PRPgtlike ltw VVGgttwisting
  ltw DPSgtyour ltw NN1gtankleltc PUNgt.
• lts n0069gt
• ltw PNPgtItltw VBZgt's ltw AV0gtalways ltw AJCgtweakerltc
  PUNgt.
• ltw PNPgtItltw VBZgt's ltw AV0gtalways ltw AJCgtweakerltc
  PUNgt.
• lt/ugt
• ltu whoPS141gt
• lts n0078gt
• ltptr tKDYLC00Bgt ltw PNPgtYou ltw VVBgtmean ltw
  PNPgtyou ltw VM0gtcaltw XX0gtn't ltw VVIgtsee
• ltptr tKDYLC00Cgt ltw PNPgtit ltw TO0gtto ltw VVIgtput
  ltw PRPgtround ltw AT0gtthe ltw NN1gtcorner
• ltw AV0gthere ltw PRPgtlike ltw DT0gtthisltc PUNgt?
• lt/ugt

10
Appendix to transcription part of speech (POS)
tags

• AJC comparative adjective
• AT0 article
• AV0 adverb (unmarked)
• CJS subordinating conjunction
• DPS possessive determiner form
• DT0 general determiner
• PNP personal pronoun
• PRP preposition (except for OF)
• PUN punctuation general mark
• TO0 infinitive marker TO
• VBZ -s form of the verb be (is)
• VDN done
• VHB forms of the verb have
• VM0 modal auxiliary verb
• VVB base form of lexical verb (except the
  infinitive)
• VVG -ing participle of lexical verb
• VV1 infinitive
• XX0 the negative NOT or NT

11
Compare Original Version with SGML (taken from
BNC Sampler Treebank)

• Original
• S0KC8001 v
• S S Hello_UH there_RL ltpausegt_NULL right_RR
  ,_YCOM and_CC N what_DDQ
• N do_VD0 N you_PPY NV want_VVI V ,_YCOM N
  some_DD mushrooms_NN2
• NS ltpausegt_NULL SN which_DDQ mushrooms_NN2
  N shall_VM N
• we_PPIS2 NV have_VHI VN my_APPGE love_NN1
  NS ?_YQUE S
• SGML
• ltS n"001"gt ltS co"first"gt ltw f"Hello" t"UH"gt
  ltw f"there" t"RL"gt ltpausegt ltw f"right" t"RR"gt
  ltc f"," t"YCOM"gt ltw f"and" t"CC"gt ltNgt ltw
  f"what" t"DDQ"gt lt/Ngt ltw f"do" t"VD0"gt ltNgt ltw
  f"you" t"PPY"gt lt/NgtltVgt ltw f"want" t"VVI"gt
  lt/Vgt ltc f"," t"YCOM"gt ltNgt ltw f"some" t"DD"gt
  ltw f"mushrooms" t"NN2"gt lt/Ngtlt/Sgt ltpausegt ltS
  co"subs"gtltNgt ltw f"which" t"DDQ"gt ltw
  f"mushrooms" t"NN2"gt lt/Ngt ltw f"shall" t"VM"gt
  ltNgt ltw f"we" t"PPIS2"gt lt/NgtltVgt ltw f"have"
  t"VHI"gt lt/VgtltNgt ltw f"my" t"APPGE"gt ltw f"love"
  t"NN1"gt lt/Ngtlt/Sgt ltc f"?" t"YQUE"gt lt/Sgt

12
How can the BNC be used to analyse British
English?

• Examples
• by comparing various channels and text-types
• by comparing speech and writing
• by comparing conversation and task-orientated
  speech
• by comparing imaginative and informative writing

13
Examples from the frequency list

• figures are given per million words
• antonyms good 1,276 vs. bad 264
• weekdays Sunday 93 vs. Tuesday 36
• living creatures dog 124 vs. fly 16
• weather wind 85 vs. sunshine 13
• sports football 67 vs. sailing 9
• colours black 226 vs. purple 11
• most frequent common noun time 1,833
• most frequent verb be 42,277
• most frequent determiner the 61,847

14
Errors and disadvantages of the BNC

• already existing typing mistakes in the texts
  provided
• typing errors made by transcribers
• sampling errors
• encoding errors
• tagging errors
• quotations in other language may lead to
  confusion (German die has nothing to do with
  mortality)
• corpus may be biased due to buzzwords at the
  time

15
Sources

• http//www.natcorp.ox.ac.uk/
• BNC Sampler
• Kennedy, Graeme. An Introduction to Corpus
  Linguistics. London Longman, 1998.
• Leech, Geoffrey/Greg Myers/Jenny Thomas (ed.).
  Spoken English on Computer Transcription,
  mark-up and application. New York Longman, 1995.
• Leech, Geoffrey/Paul Rayson/Andrew Wilson. Word
  frequencies in written and spoken English based
  on the British National Corpus. London Longman,
  2000.
• Aston, Guy/Lou Burnard. The BNC Handbook
  Exploring the British National Corpus with SARA.
  Edinburgh Edinburgh University Press, 1998.