Syntactic constructions in The Trondheim Wizard of Ozexperiment

1
Syntactic constructions in The Trondheim Wizard
of Oz-experiment

• 22.04.2006, Delmenhorst
• PhD student Heidi Brøseth
• Department of Language and Communication Studies,
  
• Norwegian University of Science and Technology
• broseth_at_hf.ntnu.no

2
Background

• Modality written HCI
• Thompson (1980), Reilly (1987), Jönsson/Dahlback
  (1988)
• Misspelling, fragmentary input, treatment of
  ellipsis (Reilly)
• Telegraphic style The syntax is simple and
  within current parser technology (Reilly)
• Modality spoken HCI
• Richards/Underwood (1985), Woffitt et al (1997)
• Politeness and regularity (Richards/Underwood)
• Turn taking, repair strategies, hesitation items
  (Woffitt et al)
• What constructions do we find in spoken HCI?
• Are the syntax found in written HCI similar to
  written HCI?
• Cultural differences? (Cameron 2001, Morel 1987)

3
(No Transcript)
4
Turn types in TWOZ

• Fragment turn (78 occurr.)
• PP, NP, ADV
• Main clause turn (303 occurr.)
• Main clause
• Main clause 1 or more subordinate clause(s)
• Co-ordinated clauses
• Complex turn (50 occurr.)
• Several main clauses
• Several main clauses 1 or more subordinate
  clause(s)
• Backchannel turn (22 occurr.)
• Mmm, yes, thanks, ok, ah

5
(No Transcript)
6
Main clauses in the TWOZ
7
Main clauses in the TWOZ, cont.

• Main clauses are the most frequent clause type in
  the material both in the opening sequence and as
  an answer to the dialog system's questions
• Jeg vil reise fra kattemskogen. ("I will travel
  from Kattemskogen.")
• Interrogative clauses can be divided into two
  main groups
• Når går neste buss fra Stavset til Nidarvoll?
• "When leaves the next bus from Stavset to
  Nidarvoll"
• Hvilken buss går fra Ankers gate til Nordre gate
  før ni i morgen?
• "Which bus leaves from Ankers gate to Nordre gate
  before nine o'clock tomorrow?"
• Polar questions occur rarly in the opening
  sequence, and never as the initial clause
• Kan du hjelpe meg med det? ("Can you help me with
  that")

8
Main clauses in TWOZ, cont.

• The topicalized constructions are rare (22
  occurr.) and they consist mostly on one-syllable
  constituents så, da, takk. (then, then, thanks)
• Da må jeg være på City Syd klokka syv i kveld
• "I must then be at City Syd at seven o'clock
  tonight"
• Imperatives 9 occurrences, Avslutt (Terminate)
  are a direct response to the dialog systems
  utterance in the first experiment. When the
  dialog system's utterance was altered, the
  imperative disappeared. Only 3 imperatives in the
  remaining material.

9
Main clauses in TWOZ, cont.

• Conditional clauses (only protasis) Utilised to
  suggest a different scenario than the one
  suggested in the user's previous turn. Never
  found in the opening sequence.
• U Vi skal på uka i morgen kveld har du noen
  nattbuss fra uka til Huseby?
• "We're going to Uka tomorrow night. Do you have
  any late night bus from Uka to Huseby?"
• DS Jeg har ingen informasjon om ruter på denne
  datoen
• "I have no information about bus schedules on
  this date."
• U Hvis vi tar nattbuss !ikveld da fra Uka til
  Huseby?
• "If we take a night bus tonight then from Uka to
  Huseby?"

10
Subordinated clauses
11
Observations

• The "lack" of intricate syntactic structures
  contradicts the assertions made by Halliday
  (1986) and Miller/Weinert (1998). They claim that
  spoken language have a lower lexical density, and
  more elaborated grammatical structures than
  written language.
• Subordination is not frequent in the TWOZ data.
  Supports the finding in Miller/Weinert that
  rapports 14.5 subordinate clauses in a spoken
  corpus when the speakers have no eye-contact.
• "The lack of non-verbal communication is
  accompagnied by an increase in the number of
  clauses but also a decrease in the number of
  subordinate clauses (Miller/Weinert, 199890-1)

12
Observations

• The TWOZ shows a lower rate of repetitions, false
  starts and corrections than for instance
  rapported by Heeman/Allan (1997)
• Heeman/Allan (1997) 2396 repairs in 6161 user
  turns (38.9)
• TWOZ 38 repairs in 453 user turns (8.4)
• Elliptic structures Few ellipsis (31
  occurrences) 90.3 of these are ellipsis that
  also are allowed or occur in written Norwegian.
• These observations do not follow the assertions
  of written HCI made by Reilly (1987)

13
Summing up

• The syntactic constructions in the TWOZ are
  homogeneous
• The syntactic constructions in the TWOZ are
  complete sentences
• The TWOZ material is in some respect similar to
  findings in other spoken corpora (low frequency
  of subordination), while not similar to other
  aspects of spoken language, for instance,
  grammatical intricacy, repetition, corrections.
• Maynor (1994) investigation of computer-mediated
  talk "written speech"
• TWOZ "spoken writing"?

14
Conclusion

• There is not a large amount of fragmented syntax,
  elliptic structures or false starts/corrections
  in spoken HCI, contrary to written HCI as
  rapported by Reilly (1987).
• If grammatical sentences and a limited array of
  sentences equals simple syntax, then Reilly
  (1987) is correct.

15
References

• Dahlbäck, N. (1995) "Kinds of Agents and Types of
  Dialogues" in Proceedings of the Ninth Twente
  Workshop on Language Technology, Universiteit
  Twente, Enschede, The Netherlands.
• Dahlbäck, N. (1997) "Towards a Dialogue Taxonomy"
  in Dialogue Processing in Spoken Language
  Systems, Elisabeth Maier, Marion Mast, Susann
  LuperFoy (Eds.). Springer Verlag Series
  LNAI-Lecture Notes in Artificial Intelligence
  1236.
• Cameron, D. (2001) Working with spoken discourse,
  SAGE Publication Ltd, London.
• Halliday, M.A.K. (1985) Spoken and written
  language (2nd edition), Oxford University Press,
  Oxford.
• Heeman, P./ Allen, J. (1997) Intonational
  Boundaries, Speech Repairs, and Discourse
  Markers Modelling Spoken Dialog,'' In
  Proceedings of the 35th Annual Meeting of the
  Association for Computational Linguistics
  (ACL-97), Madrid, July 1997.
• Jönsson. A./Dahlbäck, N. (1988) "Talking to a
  Computer is not like Talking to Your Best Friend
  in Proceedings of The first Scandinavian
  Conference on Artificial Intelligence, Tromsø,
  Norway, March 9-11.
• Maynor, N. (1994) "The language of Electronic
  Mail Written Speech? in Publication of the
  American Dialect Society 7848-54.
• Miller, J./Weinert, R. (1998) Spontaneous spoken
  language Syntax and Discourse, Clarendon Press,
  Oxford.
• Reilly, A. (1987) "Ill-formedness and
  miscommunication in person-machine dialogue",
  Information and Software Technology 2969-74.
• Richards, M.A./Underwood, K.M. (1984) How should
  people and computers speak to each other? in
  proceedings INTERACT84, pp. 215-218.
• Rubin, A. (1980) "A theoretical taxonomy of the
  differences between oral and written language" in
  Theoretical Issues in Reading Comprehension,
  Spiro, R.J., Bruce, B.B., and Brewer, W.F.
  (eds.), Erlbaum.
• Thomson, B.H. (1980) Linguistic analysis of
  natural language communication with computers in
  Proceedings of the 3rd International Conference
  on Computational Linguistics, Tokyo.
• Wooffitt et al. (1997) Humans, Computers and
  Wizards analysing human (simulated) computer
  interaction, Routledge, London.