Generating Subsequent Reference in Shared Visual Scenes: Computation vs Re-Use

Proceedings of the 2011 Conference on Empirical Methods in Natural Language Processing, pages 1158–1167, Edinburgh, Scotland, UK, July 27–31, 2011. c⃝2011 Association for Computational Linguistics Generating Subsequent Reference in Shared Visual Scenes: Computation vs. Re-Use Jette Viethen1,2 jette.viethen@mq.edu.au 1TiCC University of Tilburg Tilburg, The Netherlands Robert Dale2 robert.dale@mq.edu.au 2Centre for Language Technology Macquarie University Sydney, Australia Markus Guhe3 m.guhe@ed.ac.uk 3School of Informatics University of Edinburgh Edinburgh, UK Abstract Traditional computational approaches to re- ferring expression generation operate in a de- liberate manner, choosing the attributes to be included on the basis of their ability to dis- tinguish the intended referent from its dis- tractors. However, work in psycholinguis- tics suggests that speakers align their refer- ring expressions with those used previously in the discourse, implying less deliberate choice and more subconscious reuse. This raises the question as to which is a more accurate char- acterisation of what people do. Using a cor- pus of dialogues containing 16,358 referring expressions, we explore this question via the generation of subsequent references in shared visual scenes. We use a machine learning ap- proach to referring expression generation and demonstrate that incorporating features that correspond to the computational tradition does not match human referring behaviour as well as using features corresponding to the process of alignment. The results support the view that the traditional model of referring expression generation that is widely assumed in work on natural language generation may not in fact be correct; our analysis may also help explain the oft-observed redundancy found in human- produced referring expressions. 1 Introduction Computational work on referring expression genera- tion (REG) has an extensive history, and a wide vari- ety of algorithms have been proposed, dealing with various facets of what is recognised to be a com- plex problem. Almost all of this work sees the task as being concerned with choosing those attributes of an intended referent that distinguish it from the other entities with which it might be confused (see, for example, Dale (1989), Dale and Reiter (1995), Krahmer et al. (2003), van Deemter and Krahmer (2007), Gardent and Striegnitz (2007)). Indepen- dently, an alternative way of thinking about refer- ence has arisen within the psycholinguistics com- munity: there is now a long tradition of work that explores how a dialogue participant’s forms of ref- erence are influenced by those previously used for a given entity. Most recently, this line of work has been discussed in terms of the notions of alignment (Pickering and Garrod, 2004) and conceptual pacts (Clark and Wilkes-Gibbs, 1986; Brennan and Clark, 1996). We suspect that neither approach tells the full story, and so we are interested in exploring whether the two perspectives should be integrated. Using a large corpus of referring expressions in task-oriented dialogues, this paper presents a machine learning approach that allows us to combine features corre- sponding to the two perspectives. Our results show that models based on the alignment perspective out- perform models based on traditional REG considera- tions, as well as a number …

Content truncated to 500 words