Entrainment on the Move and in the Lab: The Walking Around Corpus

Entrainment on the Move and in the Lab: The Walking Around Corpus Susan E. Brennan (susan.brennan@stonybrook.edu) Department of Psychology, Stony Brook University, Stony Brook, NY 11794-2500 USA Katharina S. Schuhmann (katharina.schuhmann@stonybrook.edu) Department of Linguistics, Stony Brook University, Stony Brook, NY 11794-4376 USA Karla M. Batres (karla.batres@stonybrook.edu) Department of Psychology, Stony Brook University, Stony Brook, NY 11794-2500 USA Abstract We examined lexical choice and variability in referring expressions during direction-giving to pedestrians. The Walking Around Corpus comprises an experimentally parameterized collection of spontaneous spoken dialogues produced by 36 pairs of people communicating by mobile telephone; it provides both a testbed for lexical entrainment “in the wild” as well as a resource for pedestrian navigation applications. A stationary partner (the Giver) directed a mobile partner (the Follower) to walk about 1.8 miles, to 18 destinations on a medium-sized campus. Givers viewed a map marked with target destinations, labels, and photos. Followers carried a cell phone with GPS and a digital camera in order to photograph the destinations they visited; Givers monitored Followers’ progress as a cursor on a map display. Immediately after the navigation task, Followers returned to the lab and both were tested individually on their spatial ability and memory for the destinations. Next, the Experimenter attempted to interfere with any conceptual precedents they had established by giving Followers printed copies of the photos they had just taken and prompting them (sometimes with competing labels) to identify each destination. Finally, each pair participated in 6 rounds of a more traditional referential communication lab task to repeatedly match duplicate copies of the Follower’s pictures of the destinations. Results include significant rates of lexical entrainment, evidence for partner-specific conceptual pacts, and that joint navigation efficiency is affected by direction- givers’ spatial ability. The Walking Around Corpus is available to the research community. Keywords: Referential communication; conceptual pacts; entrainment; collaboration; mobile communication; GPS applications; pedestrian navigation; spatial cognition. Introduction Speakers make many expressive choices in conversation, leading to enormous variability in spontaneous speech. These choices emerge not only from their individual proclivities and the availability of words in memory, but also from collaboration with conversational partners (Clark & Wilkes-Gibbs, 1986). There is much less variation within than between conversations; speakers are more likely to continue using the same term with the same addressee than with a new addressee (Brennan & Clark, 1996). Studies of referential communication in laboratory settings show that lexical choice is influenced not only by precedent (what a speaker has said previously) and strength of precedent (e.g., how often a referent has been discussed), but also by partner (whether the precedent was established with the current partner or a different one) but that it can be maintained via interactive cues provided during grounding (ibid). We have proposed that lexical entrainment marks two speakers’ beliefs that they are referring to the same thing; in fact, breaking a conceptual pact, such as hearing a speaker inexplicably abandon an entrained-upon expression for a new expression when referring to the same referent, results in slower comprehension than hearing the same new expression from a different (new) partner (Metzing & Brennan, 2003; Matthews, Lieven, & Tomasello, 2010). A conceptual pact does not reflect a rigid mapping between expression and referent, but emerges instead from a flexible, temporary agreement arrived at with an interactive partner to take a particular perspective on the referent; as such, it is highly dependent on context. Although there is evidence that the partner associated with a conceptual pact is represented by a cue in memory (Horton, 2007; Horton & Gerrig, 2005), this does not pre-empt an expression and referent from being easily associated with a different partner (Barr & Keysar, 2002; Brennan & Clark, 1996), nor does having an existing conceptual pact with a partner inhibit a new expression-referent association from being encoded with that partner when the pragmatic context changes (ibid). Lexical entrainment has been demonstrated in adults (Brennan & Clark, 1996; Clark & Wilkes-Gibbs, 1986; Garrod & Anderson, 1986) and in children (Matthews et al., 2010); in experts and in novices (Isaacs & Clark, 1987); in native and native/non-native conversations (Bortfeld & Brennan, 1997); and for lexicalized and non-lexicalized referents (Gergle, Kraut, & Fussell, 2013; Schober & Carstensen, 2010). It has also been measured in statistical analyses of the transcripts of speech corpora (e.g., Nenkova, Gravano, & Hirschberg, 2008’s studies of the Switchboard and Columbia Games corpora, and Stoyanchev & Stent, 2009’s from Let’s Go, a real-world bus information system). In laboratory studies of referential communication and lexical choice, we and others have demonstrated effects of psycholinguistic processing in interactive dialogue contexts,