Dundee Is anaphoric reference cooperative ?

Two experiments investigated whether the choice of anaphoric expression is affected by the presence of an addressee. Following a context sentence and visual scene, participants described a target scene that required anaphoric reference. They described the scene either to an addressee (Experiment 1) or without an addressee (Experiment 2). When an addressee was present in the task, participants used more pronouns and fewer repeated noun phrases when the referent was the grammatical subject in the context sentence and no competitor was mentioned. They used fewer pronouns and more repeated noun phrases when a visual competitor was present on the scene than when there was no visual competitor. In the absence of an addressee, linguistic context effects were the same as when an addressee was present, but the visual effect of the competitor disappeared. We conclude that visual salience effects are due to adjustments that speakers make when they produce reference for an addressee, whereas linguistic salience effects appear whether or not speakers have addressees. Is anaphoric reference cooperative 3 Is anaphoric reference cooperative? One of the main concerns in discourse and dialogue studies has been speakers' level of cooperation with their addressees when referring to various discourse elements under discussion (e.g., Ariel, 1990; Bard et al., 2000; Brennan, 1995; Brennan & Clark, 1996; Clark & Murphy, 1982; Dell & Brown, 1991; Galati & Brennan, 2006; Horton & Gerrig, 2002; Horton & Keysar, 1996; Keysar et al., 2000). Speakers can use different referring expressions when they establish a link between different mentions of the same entity in the discourse. The linguistic form of these expressions varies with respect to how explicitly they are linked to their antecedents in the prior discourse (e.g., referring back to a footballer as 'the footballer', 'the player', 'Wayne Rooney' or 'he'). An ideal dialogue can be claimed to follow a cooperative principle (Grice, 1975) according to which speakers provide information that makes the interpretation of the utterances clear to their addressees. Whether the linguistic form that speakers choose for their referring expression can be seen to reflect this cooperative principle has been one of the central issues in studies of dialogue. In the present study, we focused on anaphoric expressions, that is, expressions (e.g., pronouns) that refer back to a recently mentioned constituent and investigated whether the form of these expressions differed when speakers completed a picture story to an addressee and when they completed the same picture story without an addressee being present. The idea that speakers adapt to their addressees' needs and knowledge in their contributions to the on-going dialogue has been termed audience design (e.g., Clark & Murphy, 1982; Galati & Brennan, 2010; Horton & Gerrig, 2002). According to collaborative models of discourse (e.g., Brennan & Clark, 1996; Hanna & Tanenhaus, 2004; Heller, Gorman, & Tanenhaus, 2012; Horton & Gerrig, 2002, 2005; Lockridge & Brennan, 2002; Metzing & Brennan, 2003) the information that interlocutors share or do not share at a Is anaphoric reference cooperative 4 particular point in the dialogue determines what kind of expressions speakers use for their addressees; that is, speakers are assumed to design their utterances with their addressees in mind and these partner-specific effects in communication emerge early in production processes (see Brennan & Hanna, 2009 for a review). While some discourse theories assume audience design, others claim that speakers may not necessarily construct their utterances completely with their listeners in mind but might have a more self-centred point of departure and only later adjust to the needs of their listeners (e.g., Horton & Keysar, 1996; Keysar, Barr, & Horton, 1998). A number of studies (e.g., Brown & Dell, 1987; Ferreira & Dell, 2000; Horton & Keysar, 1996; Wardlow Lane & Ferreira, 2008; Wardlow Lane et al., 2006) suggest that speakers produce language highly automatically using a production system that is less sensitive to the communicative needs of the addressees than what is assumed by collaborative models. Considering the listeners' perspective by keeping track of what has been said in the preceding context and what is known to the listeners on the basis of their general knowledge might be a time-consuming process that places high demands on speakers' memory and attention (e.g., Pickering & Garrod, 2004). Also, in an on-going conversation speakers may have very little time to plan their utterances in advance because speaker-listener switches are often done with little or no time gap at all (e,g., Beattie, 1981; Clark, 1996; Jefferson, 1972). Horton and Keysar (1996), for example, showed that speakers did not consider listeners' perspective under time pressure as often as they did when there were no time constraints. Most researchers agree that audience design occurs but disagree on when during the production process it occurs and under what circumstances. Most often, research in interactive dialogue has studied audience design in initial reference and in conditions in which speakers refer to the same object on different occasions or with different addressees. Studies have provided demonstrations of lexical entrainment; that is, when two interlocutors repeatedly Is anaphoric reference cooperative 5 refer to the same object, they tend to start using the same terms (e.g., Brennan & Clark, 1996; Garrod & Anderson, 1987; Van Der Wege, 2009). In these experiments, participants referred to objects in a context with another object from the same category (e.g., Brennan & Clark, 1996) or referred to a similar object that differed in size or shape (e.g, Brown-Schmidt & Tanenhaus, 2006; Wardlow & Ferreira, 2008) or repeatedly referred to same objects when they interacted with new and old conversational partners (e.g., Brennan & Clark, 1996; Horton & Gerrig, 2005). In other studies, participants described routes on maps that differed for the speaker and the addressee or they were engaged in cooperative maze games (e.g., Garrod & Anderson, 1987) where they described their locations in the maze to each other. What has not been systematically investigated in many of the studies is anaphoric reference. Anaphoric reference indicates that a noun phrase or a pronoun is referentially equivalent to a previously mentioned person or item (e.g., A footballer hit a supporteri. The supporteri fell. or The footballeri fell. Hei was tackled from behind.). A common assumption in theories of reference (e.g., Ariel, 1990, Chafe, 1994; Givón, 1983) is that the anaphoric expressions speakers choose when they refer to elements in the prior context signal how salient or accessible the referent is in its context. Language users produce reduced referring expressions such as pronouns when they refer to a highly accessible referent and use more explicit expressions such as names and noun phrases to refer to less salient referents; that is, the more accessible a referent is in the context the less information is needed in the anaphoric expression. When speakers use an unstressed pronoun (e.g., he) to refer to an individual (e.g., a footballer) they presumably refer to the most salient individual in the prior context whereas they use more elaborate expressions (e.g., the footballer) to refer to less accessible discourse entities. In fact, it has been shown that repeating explicit referring expressions for highly accessible referents can lead to processing difficulty (Gordon, Grosz, & Gilliom, 1993). Is anaphoric reference cooperative 6 Research has shown that various discourse factors make referents salient for language users and affect language producers' choice of anaphoric expressions. For referents that are the grammatical subject or the first-mentioned entity, most often the topic in the preceding discourse, language producers tend to use more pronouns and fewer repeated names or definite noun phrases than for referents that are not subjects or first-mentioned entities in sentences (e.g., Arnold, 2001; Fletcher, 1984; Stevenson, Crawley, & Kleinman, 1994). They also tend to use more pronouns for animate than inanimate referents (Fukumura & Van Gompel, 2011) and they produce less elaborate expressions when only a single referent is present in the preceding context than when another referent is also mentioned (e.g., Arnold & Griffin, 2007; Fukumura, Van Gompel, & Pickering, 2010). Furthermore, corpus studies show that language producers use shorter expressions for entities that have been recently mentioned than for entities that were established earlier in the context (e.g., Ariel, 1990; Arnold, Bennetto, & Diehl, 2009; Givón, 1983). However, most studies have not investigated to what extent speakers’ choice of anaphoric form is driven by the perceived needs of their addressee. Do speakers choose anaphoric expressions based on how accessible they believe that they are for their listeners or does the choice of the linguistic form merely reflect how available the referents are for the speakers themselves? Often, however, the speaker's point of departure does not differ from that of the adressees involved in the same conversation. If the speaker and the addressee share the same discourse context, referents that are highly accessible in the speakers' discourse representation are likely to be as accessible in the addressees' discourse representation. Therefore, when choosing referential expressions speakers might use the accessability of the referents that comes from their own internal representation of discourse and not necessarily from the perceived needs of the addressee. For example, Fukumura and Van Gompel (2012) manipulated whether speakers and their addressees shared the immediately preceding Is anaphoric reference cooperative 7 linguistic context in which either a referent or a competing chararcter was mentioned. The study showed that speakers’ choice of referring expressions was determined by how accessible the referent was for the speakers rather than how accessible it was for the addressees. In the present study, we examined whether and to what extent the anaphoric expressions that speakers produce differ depending on whether the speakers interact with an addressee or not. Many linguistic models of discourse representation (e.g., Asher & Lascarides, 2003; Grosz, Joshi, & Weinstein, 1995; Hankamer & Sag, 1976; Heim, 1982; Kamp & Reyle, 1993; Webber, 1979) make the implicit assumption that speakers' discourse representations are primarily anchored to the preceding linguistic context; anaphoric expressions are syntactically and semantically controlled by their linguistic antecedents within the context. Although these models assume that a reference can also be pragmatically controlled without an explicitly expressed antecedent (e.g., derived from the context by lexical inference as in Keith drove to London. The car was old and rusty where the referent the car is inferred from the verb drive) and non-linguistic information can affect reference in discourse (e.g., Hankamer & Sag, 1976), their architectures generally only include elements that are linguistically mentioned. They do not capture non-linguistic elements that are part of external situations or need extensive inference. One such external source of information that has received considerable attention in experimental research is the visual context. Several studies (e.g., Brown-Schmidt & Tanenhaus, 2006; Tanenhaus, Spivey Knowlton, Eberhard, & Sedivy 1995; Wardlow Lane & Ferreira, 2008) have shown that language processing in general is affected by what is visually present in the discourse context and that the linguistic forms speakers use are influenced by what they can see, even in cases in which the visual information is not shared by their addressees. Three studies (Arnold & Griffin, 2007; Fukumura, Van Gompel, & Pickering, Is anaphoric reference cooperative 8 2010; Vogels, Krahmer, & Maes, 2012) have investigated whether speakers' choice of anaphoric expressions is also affected by visual context. Arnold and Griffin (2007) investigated the use of pronouns and repeated names in a story-telling task. Participants first heard a context sentence that described a stimulus picture showing one or two characters (e.g., Mickey went for a walk in the hills / Mickey went for a walk with Daisy in the hills). Next, they described a target picture in which one of the characters (e.g., Mickey) was engaged in an action. In the single-character condition only one character was mentioned (e.g., Mickey went for a walk in the hills) and was present in both pictures. In the other conditions, a competing character of a different gender (Daisy) was mentioned in the context sentence and was present both in the stimulus picture and the target picture (the two-character condition), or the competitor (Daisy) was mentioned in the context sentence and was shown in the stimulus picture but was no longer present in the target picture (the two/one-character condition). Arnold and Griffin found a higher mean use of pronouns than repeated names in the single character condition whereas pronoun use was much lower in the other two conditions which did not differ. Pronoun use was significantly lower in the two/one-character conditions than in the single-character condition although the target picture was exactly the same in both conditions. According to Arnold and Griffin, the speakers' choice of referring expressions was affected by the presence of a competing character in the discourse representation rather than what was visually present in the target picture. But Arnold and Griffin's (2007) results may suggest that the visual context per se did not affect the choice of referring expressions. Pronoun use was similar when the competitor was present in the target picture (the two-character condition) to when it was not (the two/one-character condition) suggesting that the choice of referring expressions was only affected by whether the context sentence mentioned the competing character or not. Participants produced fewer pronouns when the competitor was mentioned in the context Is anaphoric reference cooperative 9 sentence (the two-character condition and the two/one-character condition) than when it was not (the single character condition). Vogels et al. (2012) used a story completion task in which participants saw two characters in a picture and heard two sentences that mentioned both characters. Next, the participants saw a target picture in which one of the characters performed an action and were asked to describe the picture. Vogels et al. manipulated the visual salience of the characters in the target picture by placing one of the characters in the foreground and the other in the background. The participants were more likely to start their description by referring to the visually salient character in subject position than the visually non-salient character. However, visual salience did not affect the choice of the anaphoric expressions. The relative proportions of pronouns and definite noun phrases were similar for the visually salient character and nonsalient character. In contrast, Fukumura et al. (2010) did find an effect of visual presence on anaphoric reference. In this study, speakers first read aloud a context sentence (e.g., The pirate's carpet had been cleaned / The pirate's carpet had been cleaned by a prince) and saw a picture in which there were either one (the pirate) or two (the pirate and a prince) characters. Both the linguistic mention and visual presence of a competitor were manipulated; the competitor (a prince) was either mentioned or not mentioned in the context sentence, and the competitor was either present or absent in both the context picture and the target picture. The speakers' task was to describe the target picture in which the main character (the pirate) performed an action to an addressee who carried out the action using toy characters. Speakers produced pronouns (relative to repeated noun phrases) for the target character less often when the competitor was mentioned in the context sentence than when no competitor was mentioned. Moreover, the speakers produced fewer pronouns when the competitor was visually present than when it was absent. This effect also occured in a further experiment in which the referent Is anaphoric reference cooperative 10 and competitor had different genders (e.g., a pirate and a princess) and pronominal reference was unambiguous. Fukumura et al. concluded that both the linguistic mention and the visual presence of a competing character reduce the salience of the referent resulting in more repeated nouns and fewer pronouns. The results from Arnold and Griffin (2007) and Fukumura et al. (2010) showed that the linguistic mention of a competitor reduces the prominence of the referent for speakers. Fukumura et al. also showed that the presence of a competitor, even when it was not mentioned, resulted in fewer pronouns and more repeated noun phrases, whereas there was no unequivocal effect of visual context in the Arnold and Griffin and Vogel et al. (2012) studies. Speakers in the Arnold and Griffin and Vogel et al. studies had no real addressee so the use of reference in these studies may well have reflected speakers’ own discourse representation; that is, what is easiest for speakers. This representation may not include elements from nonlinguistic context but could be based on linguistic information only. But in communication with addressees, speakers might also take into account the non-linguistic information and use expressions that would better distinguish between potential references if they believed that these might benefit communication. This would suggest that there are two separate sources, the linguistic and the visual context, that affect speakers' discourse model. The linguistic context appears to be unaffected by the presence of an addressee whereas visual context effects could be the result of the speakers' awareness of the non-linguistic environment that they share with their listeners. Interestingly, Wolter, Skovbroten Gorman, and Tanenhaus (2011) demonstrated that in language comprehension, linguistic and non-linguistic information made separate contributions to reference resolution. Their participants moved objects in a visual display according to instructions while their eye movements were recorded. The experimental set contained two objects that contrasted in size (e.g., a big candle and a small candle) among Is anaphoric reference cooperative 11 other objects. Participants heard a sequence of two sentences that asked them to move the objects in the set (e.g., Put the big candle above the heart. Now, put the small... or Put the object in the yellow square above the triangle. Now, put the small...). When the first part of the instructions contained a prenominal scalar adjective (e.g., big) that contrasted the objects in the set, the participants, upon hearing the adjective (e.g., small), looked more often at the other member in the set (the small candle) relative to a distractor (e.g., a small tie) than when the first part of the instructions referred to the same object (e.g., the bigger candle in the set of two candles) but did not contain the scalar adjective (e.g., Put the object in the yellow square...). Instructions that did not contain the size description did not lead to an anticipation effect on the other member of the set. Wolter et al. argued for a language processing model in which linguistic mention is more prominent and differentiated from other types of information that affect reference resolution in discourse. This model is in line with discourse representation theories that implicitly assume that discourse representations are primarily constructed on the basis of the linguistic input (e.g., Asher & Lascarides, 2003; Grosz, Joshi, & Weinstein, 1995; Heim, 1982; Kamp & Reyle, 1993). In the present study, we asked whether linguistic information and visual context influence speakers' choice of anaphoric expressions differently depending on whether they interact with addressees or not. To test this, we manipulated the linguistic mention and the visual presence of a character when speakers produced referring expressions either for an addressee (Experiment 1) or without addressees (Experiment 2). The comparison of the two experiments allowed us to investigate whether speakers use anaphoric expressions that clearly and unambiguously identify the referent more often in a task with addressees who benefit from the descriptions than in a task that was otherwise the same but no addressee was present. In our study, we used a story-telling task (e.g., Arnold & Griffin, 2007). Participants first saw a picture of one or two cartoon-like characters involved in an activity (see Fig. 1) Is anaphoric reference cooperative 12 and heard a context sentence (e.g., A supporter kicks a footballer) that described the picture. Next, they saw a follow-up picture (Fig. 2) in which one of the characters performed an action (e.g., throwing away a bottle in anger). The participants' task was to describe the follow-up picture. We analysed how participants referred to the character in the target picture (e.g., The footballer/He throws away a bottle). In Experiment 1, participants described the picture to an addressee whereas in Experiment 2, there was no addressee present and no potential addressee was mentioned in the instructions. The experimental manipulation of the context sentence and the first picture included four conditions (Figs. 1a-d). We investigated the effect of grammatical role/order-of-mention on the choice of anaphor by comparing a condition in which the character in the target picture (Fig. 2) was a grammatical object and the second-mentioned character in the preceding context sentence (Fig. 1a: object antecedent competitor mentioned and visible) to a condition in which the target character was a subject and mentioned first (Fig. 1b: subject antecedent competitor mentioned and visible). In both conditions, a competitor (e.g., a supporter) was mentioned in the context sentence and visually present in the first picture. On the basis of previous research (e.g., Fletcher, 1984; Fukumura & Van Gompel, 2010; Stevenson, Crawley, & Kleinman, 1994), we expected that when describing the target picture (Fig. 2) the participants would use more pronouns and fewer repeated noun phrases when they referrred back to the subject (Fig. 1b) than to the object (Fig. 1a) in the context sentence. Also, assuming that preceding linguistic information is used regardless of the presence of an addressee, we expected that the effect of grammatical role or order of mention (topic vs. nontopic) would not differ between the experiments. The effect of linguistic mention on reference to the target character was investigated by comparing the condition in which the target character was the subject of the context sentence and a competitor was mentioned and visually present (Fig. 1b: subject antecedent Is anaphoric reference cooperative 13 competitor mentioned and visible) to a condition in which only the target character was mentioned in the context sentence but the competitor was visually present (Fig. 1c: subject antecedent competitor visible). Based on the results of Arnold and Griffin (2007) and Fukumura et al. (2010), we expected that participants would use fewer pronouns when the competitor was mentioned (Fig. 1b) than when it was not (Fig. 1c). If anaphoric usage is unaffected by addressee presence, then the between-experiment addressee manipulation should have no effect on anaphor choice. Finally, we examined the visual context effect by comparing the two conditions in which only the target character was linguistically mentioned in the context sentence (Figs. 1c and 1d). The linguistic context was exactly the same in these conditions but in one condition a competitor was visually present (Fig 1c: subject antecedent competitor visible) whereas in the other it was not (Fig. 1d: subject antecedent no competitor). If speakers' attention is drawn to the visually present character they should use fewer pronouns in the condition in which a competitor is visually present (Fig. 1c) than when no competitor is present (Fig. 1d). The question of interest here was whether the visual competitor effect would be the same in both tasks or whether it would depend on the speaker's task. That is, does the visual competitor affect the way how speakers refer to the target character only in a task that involves interaction with an addressee or does it also occur in a task without an addressee.