Lattice Theoretic Relevance in Incremental Reference Processing

While there has been substantial work on referential communication tasks in psycholinguistics, computational and formal modelling (Dale and Reiter (1995), Krahmer and Van Deemter (2012), Frank and Goodman (2012) inter alia), the element we discuss here is incremental processing. Motivated by work in incremental generation of referring expressions (Guhe, 2007; Fernández, 2013) and incremental reference resolution in NLU (Kennington and Schlangen, 2014), we present a dialogue-motivated account which models the speaker and the hearer in reference identification games. A central desideratum of an incremental account of reference identification tasks can be found in the evidence from Brennan and Schober (2001)’s experiments; namely that people reason at an incredibly time-critical level from linguistic information. They demonstrated selfrepair can speed up semantic processing (or at least object reference) where an incorrect object being partly vocalized and then repaired in the instructions (e.g. “the yell-, uh, purple square”) yields quicker response times from the onset of the target (“purple”) than in the case of the fluent instructions (“the purple square”), with little effect on accuracy. We wish to model this faculty of repair processing, and also wish to model non-local repair processing of instructions such as “From yellow down to brown – no – thats red.” (Levelt, 1989, via Ginzburg et al. (2014)), here using a syntactically simpler but illustrative alternative “the yellow square, no, purple”. We build on Hough and Purver (2014)’s integration of Knuth (2005)’s lattice-theoretic characterization of probabilistic inference to model interpretation of repaired instructions in a small reference domain.