Conclusions and Future Work

The results are approximately 60 % failures and 40 % success. Behind these figures lay the other’s image and the expectations, both from the speaker and from the machine. One of the principles that makes human communication possible is that speakers are able to adapt their speech acts to the expectations of their interlocutors (Levinson 1983). To succeed it is necessary to figure out what those expectations are in order to achieve the joint goals. Under this theoretical frame, misunderstandings are produced when there is a disagreement between interlocutors about the other’s image. In this case, the machine is programmed to expect some linguistics and pragmatic behaviours from the human that very often do not occur. On the other hand, the speaker’s expectations are not met because of the limited capabilities of the system.

In this paper, we have shown that in the C-ORAL-ROM corpus when the human speaker is aware of his/her dialog with a limited system and consequently adapts his/her discursive strategy, the conversation ends in a successful acquisition of information. On the other hand, when the human speaker adopts the same strategies, courtesy and information exchange as if he/she would be talking to another human being sharing the same code and pragmatic use, then the request for information results in a failure. The most common strategy used by the speakers who got a productive information exchange was to give in to the machine the dominant role and constraint themselves to answer in the most concrete and relevant way to the questions formulated by the automatic system. As a main conclusion, the success is highly dependent on the ability of the speaker to talk to the machine and not the other way round.

In order to improve the satisfactory exchange with the user, dialog systems should add in their design the recognition of misunderstandings and implement strategies to solve them. For instance, they should manage to recognise when a human speaker try to express a problem with his communicative interests. Obviously, this is language- dependant. Another improvement could be to insert repair sequences (Jefferson 1974), very common in human communication in order to draw interlocutor’s attention to a problem in the interaction. But probably the most practical improvement would be to clearly state that the system is not able to handle courtesy and indirect speech, and suggest to strictly follow the system instructions.

In future research we plan to analyse conversational strategies based on the sex and age variables. Our preliminary hypothesis is that men and women use different strategies to solve a misunderstanding. Similarly, young people communicate with computer systems in a different way than older people.

.

4.  References

Austin, J.L. (1962) How to do things with words (Oxford: Oxford University Press).

Cresti, E. & Moneglia, M. (eds.) (2005) C-ORAL-ROM: Integrated Reference Corpora for Spoken Romance Languages (Amsterdam: John Benjamins).

Falavigna, D. And Gretter, R. (2003) Procedures for validation of the spoken dialogue system on aligned corpora. Deliverable 7.2, C-ORAL-ROM project. Available on-line from http://lablita.dit.unifi.it/coralrom.

Gallardo, B. (1996) Análisis conversacional y pragmática del receptor (Valencia: Episteme).

Jefferson, G. (1974) Error correction as a interactional resource, Language in society, 2, 181-199.

Levinson, S. C. (1983) Pragmatics (Cambridge: Cambridge University Press). Searle, J.R. (1969): Speech Acts (Cambridge, Cambridge University Press).