Construction Repetition Reduces Information Rate in Dialogue

Proceedings of the 2nd Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics and the 12th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), pages 665–682 November 20–23, 2022. ©2022 Association for Computational Linguistics 665 Construction Repetition Reduces Information Rate in Dialogue Mario Giulianelli Institute for Logic, Language and Computation University of Amsterdam m.giulianelli@uva.nl Arabella Sinclair Department of Computing Science University of Aberdeen arabella.sinclair@abdn.ac.uk Raquel Fernández Institute for Logic, Language and Computation University of Amsterdam raquel.fernandez@uva.nl Abstract Speakers repeat constructions frequently in dialogue. Due to their peculiar information- theoretic properties, repetitions can be thought of as a strategy for cost-effective communica- tion. In this study, we focus on the repetition of lexicalised constructions—i.e., recurring multi- word units—in English open-domain spoken dialogues. We hypothesise that speakers use construction repetition to mitigate information rate, leading to an overall decrease in utter- ance information content over the course of a dialogue. We conduct a quantitative analysis, measuring the information content of construc- tions and that of their containing utterances, estimating information content with an adap- tive neural language model. We observe that construction usage lowers the information con- tent of utterances. This facilitating effect (i) in- creases throughout dialogues, (ii) is boosted by repetition, (iii) grows as a function of repetition frequency and density, and (iv) is stronger for repetitions of referential constructions. 1 Introduction The repeated use of particular configurations of structures and lexemes, constructions, is pervasive in conversational language use (Tomasello, 2003; Goldberg, 2006). Such repetition can be under- stood as a surface level signal of processes of co- ordination (Sinclair and Fernández, 2021) or ‘in- terpersonal synergy’ between conversational part- ners (Fusaroli et al., 2014). Speakers may use rep- etitions to successfully maintain common ground with their interlocutors (Brennan and Clark, 1996; Pickering and Garrod, 2004), because they are primed by their recent linguistic experience (Bock, 1986), or to avoid a costly on-the-fly search for alternative phrasings (see, e.g., Kuiper, 1995). At the same time, repetitions are also advantageous for comprehenders. Repeating a sequence of words positively reshapes expectations for those words, allowing comprehenders to process them more rapidly (for a review, see Bigand et al., 2005). As speakers are known to take into consideration both their own production cost and their addressee’s processing effort (Clark and Wilkes-Gibbs, 1986; Clark and Schaefer, 1989; Frank and Goodman, 2012), its two-sided processing advantage, as de- scribed above, makes construction repetition an efficient, cost-reducing communication strategy. In this paper, we investigate whether and how these information-theoretic properties of repeti- tions shape patterns of information rate in open- domain spoken dialogue. Information theory is the study of the conditions affecting the transmission and processing of infor- mation. To the foundations of the field belongs the noisy-channel coding theorem (Shannon, 1948), which states that for any given degree of noise in a communication channel, it is possible to communicate discrete signals nearly error-free up to a maximum information rate, the channel capacity. If speakers use the communication channel optimally, they …

Content truncated to 500 words