Alignment of Reference Frames in Dialogue

Previous research has shown that interlocutors in a dialogue align their utterances at several levels of representation. This paper reports two experiments that use a confederate-priming paradigm to examine whether interlocutors also align their spatial representations during dialogue. Experiment 1 showed a significant reference frame priming effect: Speakers tended to use the same reference frame to locate an object in a scene as the frame that they had just heard their interlocutor use. Experiment 2 demonstrated the same pattern even when the speaker’s description and their partner’s previous description involved different prepositions. Hence the effect cannot be explained in terms of lexical priming of a particular preposition. Our results are strong evidence that interlocutors in a dialogue align non-linguistic as well as linguistic representations. Research on dialogue has suggested that the traditional methods employed in psycholinguistics may not give a true, or at least complete, account of human language. The traditional approach focuses largely on monologue and involves investigating single word utterances in isolated controlled circumstances, e.g. the picture naming paradigm, or the lexical decision task. However, Clark (1996) pointed out that the natural setting for language is dialogue, and that language does not normally occur in these isolated circumstances, thus questioning the ecological validity of traditional methods. The realization of this has led to a research program into how language is used in dialogue (e.g., Clark & Wilkes-Gibbs, 1986; Horton & Keysar, 1996; Garrod & Anderson, 1987). Research in this framework has shown that interlocutors in a dialogue tend to align their utterances: Over the course of a conversation participants will come to communicate in a similar fashion to each other. This occurs at several levels of communication, including the conceptual (Garrod & Anderson, 1987), lexical (Clarke & Wilkes-Gibbs, 1986) and syntactic (Branigan, Pickering, & Cleland, 2000) levels. In these experiments, participants usually achieved alignment without resorting to overt negotiation. In the case of syntactic alignment at least, many subjects were not aware that they were aligning. Pickering and Garrod (in press) proposed a mechanism for how alignment is achieved between interlocutors. According to this theory, alignment is the basis for successful dialogue; misunderstanding occurs when alignment is not achieved. Alignment occurs when the two interlocutors employ equivalent representations at different levels, and arises from an automatic priming mechanism. This allows alignment to be achieved quickly and efficiently without reliance upon time-consuming strategies of open negotiation. Indeed, such strategies are only employed when the primitive mechanisms fail. To prevent unnecessary negotiation Pickering and Garrod suggest a second primitive mechanism that allows repair of representations when misalignment occurs; see Garrod and Pickering (2004) for a summary. Dialogue research has shown alignment of linguistic representations, but alignment is hypothesized also to occur for conceptual representations, such as those associated with object location. A speaker’s conceptual representation of where objects are located is reliant upon an overall spatial representation, which underpins the use of spatial language. In order to describe object locations effectively it is important that both interlocutors take the same perspective (Levelt, 1989) concerning the objects they are locating. For example, an addressee must understand whose left a speaker is talking about. In the same way that interlocutors align on which lexical terms should be used to describe a scene, it would be advantageous for interlocutors to align on which perspective a scene should be described from. The perspective that is taken depends upon the reference frame that is applied to a spatial representation of a scene. A reference frame is an axial co-ordinate system that defines regions extending from the origin, whose axes are labelled with directional terms. The object to be located (figure object) can then be located in relation to another object (reference object) based upon the directional axes of the reference frame. However, there are three different types of reference frame (at least in English; other languages use only two or even one; Levinson 2003) that a speaker can employ in order to locate an object: absolute, relative, and intrinsic. It is important that the addressee knows which of these the speaker is using in order to successfully understand an utterance. The absolute reference frame locates an object based upon salient, stable features of the environment, for example, the cardinal directions. The dot in Figure 1 can be described as west of the chair if the page is held horizontally with the top of the page facing north. The intrinsic reference frame locates an object based upon the directional features of the reference object. The dot in Figure 1 can be described as above the chair because it is in alignment with the top of the chair. Figure 1. The dot can be described as west using an absolute reference frame, above using an intrinsic reference frame or left using a relative reference frame. The relative reference frame locates an object in relation to the viewpoint of an observer. The axes of the reference frame are labelled based upon the features of the person upon whose viewpoint the location is based. In Figure 1 the dot would be described as left of the chair using a relative reference frame. (In many cases the relative reference frame is used from the viewpoint of a speaker or an addressee, but it can also be from a third person perspective.) The above tripartite classification of reference frames follows that proposed by Levinson (1996, 2003), and is distinct from the classification traditionally employed in the psycholinguistic literature, which identified absolute, deictic, and intrinsic reference frames, all defined on the basis of their origin. (Deictic reference frames are all reference frames with an egocentric origin.) Levinson pointed out that this traditional system is not an appropriate way to categorize reference systems because it is possible to have a non-deictic relative reference frame, such as The ball is to the right of the tree as you look at it, and a deictic intrinsic reference frame such as The ball is in front of me. When describing an object’s location, an individual has to select one of these reference frames to use in preference to either of the other two reference frames. Carlson-Radvansky and Jiang (1998) showed that reference-frame selection is achieved via inhibition of non-selected reference frames. When participants used a relative reference frame to identify an object’s location, they were slower to describe an object’s location using an intrinsic reference frame immediately afterwards. Inhibition operates not only on the endpoint of an axis, but on at least the entire axis, e.g. if left (intrinsic) is inhibited then using right or left (intrinsic) in the subsequent description will take longer than using a relative reference frame. The findings of Carlson-Radvansky and Jiang (1998) suggest that reference frames are influenced by low-level priming. However, the results do not establish whether or not this occurs during dialogue: Reaction time was used as a measure of cognitive effort in trials whereas in dialogue any effect of priming must manifest itself by a change in the person’s linguistic behaviour. Furthermore, CarlsonRadvansky and Jiang’s (1998) experiment only investigated inhibition of the endpoint of an axis and the inhibition of the axis itself. If interlocutors align reference frames, we would expect them to align the entire reference frame rather than just part of it. Therefore it is unclear whether this kind of priming is enough to cause the alignment of reference frames between interlocutors in the manner described by Pickering and Garrod (in press). In two series of experiments Schober (1993, 1995) showed that the reference frame which an individual selects is affected by their partner in a conversation. Individuals who described the location of an object to a partner who viewed the scene from a different perspective were more likely to describe the location from their partner’s perspective. When the partner queried such descriptions, they used their own perspective to describe object location. Schober concluded that interlocutors use conscious strategies to collaborate in ways of describing object location. Schober’s results suggest that interlocutors may align reference frames. However, it is not clear that this is necessarily the case. In his experiments, two participants interacted freely, allowing little control over what was said by each pair. This means that pairs of participants may be reverting to default reference frames. Furthermore, in a large proportion of trials participants located objects using terms that required no reference frames (e.g. next to, between and so on). The present work is an experimental investigation to discover whether or not interlocutors align reference frames. The investigation uses a confederate-priming paradigm (e.g. Branigan et al., 2000) where a naïve participant and a participant who is unknown to the naïve participant a confederate of the experimenter and who is following a script, communicate during the experiment. If interlocutors do align reference frames then they will use a reference frame significantly more when they have just heard an utterance using that reference frame than when they have just heard an utterance using an alternative reference frame. Alternatively interlocutors may select a reference frame based solely upon the perceptual properties of the spatial array, in which case they should be unaffected by the reference frame just used by their partner. Our experiments also set out to separate priming for reference frames from lexical priming. If priming of reference frames exists separately from lexical priming, we can expect subjects to use a reference frame significantly more if they have just heard an utterance using that reference frame even if the same spatial term is not applicable to both utterances.