Compiling a corpus of transcribed speech

1
Compiling a corpus of transcribed speech
2
Anyqs

• A corpus for classroom use in training
  interpreters
• Transcribed spontaneous speech (hard to come by)
• Understandable without much contextual
  information (standard format)
• Contemporary
• A reasonable quantity (currently 850k words)
• Basic HTML markup in official transcript
  (utterances, non-verbals)
• Easy to encode in TEI and to index with XAIRA

3
No way is this publicly available

• The BBC site contains transcripts of all Any
  Questions programmes in the last 3 years, which
  you can download freely for personal
  non-commercial use.
• But/and you cannot adapt, alter or create a
  derivative work except for your own personal,
  non-commercial use.

4
What the BBCs original looks like

• PRESENTER Jonathan DimblebyPANELLISTS Lord
  FalconerMalcolm RifkindAnne McElvoyChris
  HuhneFROM Medical Women's Federation, Central
  LondonDIMBLEBYWelcome to London where we are
  on the edge of Regent's Park at the Royal College
  of Obstetricians and Gynaecologists. Our host
  here is the Medical Women's Federation, which is
  holding its 90th anniversary conference here.
  With its origins in the late 19th Century the
  federation was in 1917 formed with an initial
  membership of 190 women doctors. Subjects at the
  top of their agenda then Medical women engaged
  in war and the contemporary challenges of
  venereal diseases, prostitution, maternity and
  infant welfare. Plus ca change. Except that today
  more than half the present crop of medical
  students are women and the federation's main aim
  is to keep women doctors active in the medical
  workforce with all that that implies for
  part-time training and child welfare.On our
  panel the former Lord Chancellor Charlie
  Falconer. Lord Falconer there have been
  scurrilous reports in some of the newspapers to
  the effect that you're not happy with your
  pension and that you want it to be doubled, it's
  52,000 a year, we can presume that you are quite
  happy yes?FALCONERI think I'd rather not talk
  about that, if you don't mind Jonathan.DIMBLEBY
  You're entirely free not to talk about that which
  suggests that it's unresolved.The former Foreign
  Secretary, Sir Malcolm Rifkind Chris Huhne who
  wants to be the next leader of the Liberal
  Democrats - do you like being the
  underdog?HUHNEI'm not sure, I think - I'm
  working on it, I'm ambitious not to be the
  underdog Jonathan.DIMBLEBYAnd Anne McElvoy,
  executive editor and columnist at the Evening
  Standard. CLAPPINGOur first question
  please.HICKSTom Hicks. Should Ian Blair
  resign?

5
Marking it up in XML

• In the Header
• Programme details (including date)
• Participants and roles
• Setting
• In the Text
• Utterance boundaries and their speakers
• Sentence boundaries (based on punctuation in
  transcript)
• Non-verbal events (e.g. clapping, laughter,
  coughs)
• Topic boundaries (i.e. new question)
• Tokenisation - s
• Pos tagging maybe some day

6
Overall document structure

• ltTEIgt
• ltteiHeadergt
• ltfileDescgt
• lttitleStmtgt
• lttitlegt
• Any questions ltdategt Date lt/dategt
• lt/titlegt
• lt/titleStmtgt
• lt/fileDescgt
• ltprofileDescgt Profile lt/profileDescgt
• lt/teiHeadergt
• lttextgt Text lt/textgt
• lt/TEIgt

7
Profile

• ltprofileDescgt
• ltparticDescgtltlistPersongt
• ltperson whoname sex f m role
• presenter questioner party
  professiongt
• ltparagt fullname lt/paragt
• lt/persongt
• lt/listPersongtlt/particDescgt
• ltsettingDescgt
• ltsettinggt wherefrom lt/settinggt
• lt/settingDescgt
• lt/profileDescgt

8
Text

• lttextgt
• ltdiv typeintrogt
• ltu whoDIMBLEBYgt
• ltsgtWelcome to London lt/sgt
• ltsgtAnd Anne McElvoy, executive editor and
  columnist at the Evening Standard. lt/sgt
• ltevent descclapping/gt
• ltsgtOur first question please.lt/sgt
• lt/ugt
• lt/divgt
• ltdiv typequestiongt
• ltu whoHICKSgt
• ltsgtTom Hicks. lt/sgt
• ltsgtShould Ian Blair resign? lt/sgt
• lt/ugt
• lt/divgt
• lt/textgt

9
Utterances / Sentences

• Role
• Lab 1438 / 4617
• Con 1225 / 4272
• Lib 701 / 2590
• Presenter 5437 / 9204
• Questioner 838 / 1582
• Other 2282 / 8480

• Sex
• Male 10096 / 24829
• Female 1817 / 5906
• Other 6 / 7

Total 11921 / 30745
10
Things to do with it (1) Politically preferred
lexis?

• Occurrences per 1000 ltsgt
• UK 293 9
• Lab 35 7
• Con 11 3
• Lib 16 6
• Other 230
• United Kingdom 67 2
• Lab 13 3
• Con 19 4
• Lib 6 2
• Other 29

• Occurrences per 1000 ltsgt
• Britain 371 12
• Lab 58 12
• Con 84 20
• Lib 31 12
• Other 198

11
Things to do with it (2) Fairly frequent
features spontaneous speech

• Agreement
• Agree (138)
• Absolutely / actually / certainly / completely /
  entirely / fully / quite / rather / totally
• Disagree (72)
• Fundamentally / profoundly

12
The advantages of a small specialised corpus

• Homogeneous
• Knowable/predictable
• Manageable numbers
• Deal with all of the data
• Less distracting

13
Another appropriate linguistic area

• Connectors
• In fact (163)
• However (131)

14
However connector or adverb?
15
And another the subjunctive in speech

• It were (128)
• As it were (104)
• If it were (20)
• I wish it were (2)

16
As it were, data
17
Women and vague language

• As it were 104
• Male 101
• Female 3

18
As it were (female speakers)
19
So are men vaguer, or do they mark vagueness
more? Is this a male politeness strategy?

• As it were
• BNC male 291
• BNC female 68
• ltugt BNC male 301205
• ltugt BNC female 306293

20
Do men and women use as it were in the same way?

• More research needed