1st Semester 2020/21: Measuring Alignment in Dialogue
- Arabella Sinclair
In dialogue, speakers will often adapt the language they use to one another. This alignment, which can happen across levels of communication (from gesture, accent, sentence structure, to word choice), is thought to be as an aid to mutual understanding, to build social rapport, or be a result of a cognitive priming process, or of implicit learning.
For example, alignment can be structural, i.e. using similar constructions as your interlocutor such as being more likely to say “X gave Y the Z” rather than “X gave the Z to Y” after hearing the first construction used. Alignment can also be lexical, in that if one speaker refers to a piece of furniture as a “sofa”, the other is more likely to use the same word, rather than “settee” or “couch” afterwards.
Quantifying this alignment involves measuring aspects of the language overlap between speakers in dialogue corpora, and whether there are local effects to this alignment (to measure priming), or whether the similarity in shared language between speakers is higher than the by-chance vocabulary repetition effects of the topic of conversation.
In this project, we will analyse alignment patterns present in some large scale conversational corpora, investigating different measures and their effects, and analysing what aspects of alignment they are capturing. In the first week, there will be two introductory sessions presenting theories of alignment and techniques to measure it in practice. In the second week students will begin to research and choose their research question specifically, while attending a practical lab session to explore some existing measures and apply them to the chosen corpora. The final two weeks will be dedicated to exploring theoretically motivated extensions of the measures discussed. In the end, the project should be documented in a report and submitted code containing the analyses. In the final week there will be a discussion session where students present their findings and compare and comment on those of the others.
Potential projects could involve:
Adding additional factors to alignment i.e. word complexity
Constraining which aspects of language are measured when analysing alignment
Choosing a certain dialogue setting where reasons for alignment may be stronger or weaker and taking this into account.
ideally, participants have taken NLP1
ability to analyse data (e.g., using Python)
Attending and completing exercises in a practical session in the second week
Report of the project + analysis code
Interactive presentation group discussion of results
Pickering & Garrod, Towards a mechanistic psychology of dialogue. Behavioural and Brain Sciences, 27:169-225, 2004. [PDF]
Brennan & Clark (1996), Conceptual pacts and lexical choice in conversation. Journal of Experimental Psychology, 22:1482–1493, 1996. [PDF]
Ferreira, V. S., & Bock, K. (2006). The functions of structural priming. Language and cognitive processes, 21(7-8), 1011–1029. 2006. [PDF]
Giles, H., & Baker, S. C. (2008). Communication Accommodation Theory. The International Encyclopedia of Communication. [PDF]