We agree with Pickering & Garrod's (P&G's) claim that models of language use must take into account the fact that production and comprehension processes are interwoven in time and interconnected as in the case of split turns. Indeed, the most basic form of language use is arguably conversation, in which interlocutors act both as producers and addressees, and each type of act involves elements of the other.
The importance of examining dialogic processes has become apparent following a surge of interest in studying language in natural settings (Pickering & Garrod Reference Pickering and Garrod2004; Trueswell & Tanenhaus Reference Trueswell and Tanenhaus2005). This interest has generated new, significant findings in conversation including development of techniques for independently quantifying and predicting the degree of coordination in conversation (Richardson et al. Reference Richardson, Dale and Kirkham2007), as well as evidence that coordinated contextual representations facilitate use of potentially ambiguous referential expressions (Brown-Schmidt & Tanenhaus Reference Brown-Schmidt and Tanenhaus2008). Similarly, experiments in noninteractive settings increasingly focus on dialogue-relevant questions such as how representations of others' mental states guide processing (Ferguson et al. Reference Ferguson, Scheepers and Sanford2010).
A central feature of language use is that it is produced and understood with respect to a particular context that constrains both what we say and how we say it. In conversation, the basic context is often assumed to be the interlocutors' common ground (Clark Reference Clark1996) with conversational efficiency increasing as common ground grows (Wilkes-Gibbs & Clark Reference Wilkes-Gibbs and Clark1992). According to P&G's proposal, context plays a key role in circumscribing prediction and, as a result, conversational efficiency.
According to P&G, listeners predict upcoming utterances using one of two mechanisms, simulation and association. Listeners use the simulation mechanism with familiar partners, and the association mechanism when they perceive themselves as being different from their partner, as in adult-child conversations, and in reading, where the addressee does not typically speak. In what follows, we argue that with both familiar and unfamiliar partners, and in talking and reading alike, language users make sophisticated predictions about future language use, based on available contextual information.
P&G argue that an addressee using the simulation mechanism will predict an utterance using an inverse model and contextual information – a prediction about what the speaker would say. P&G do not specify the details of this contextual information. However, to predict what the speaker would say requires assessing the speaker's context, where context must be broadly defined to include both local perceptual information as well as historical information about the person's dialect and past experience. After all, depending on a person's age and regional dialect, a given semantic meaning might be expressed as great, rad, or wicked. Similarly, in cases where a potential alternative referent is occluded from the speaker's but not the addressee's view (e.g., the speaker sees one cup whereas the addressee sees two), the context would predict different referential expressions depending on which perspective was used. We read P&G as saying that in each of these cases, on the simulation mechanism, the addressee would predict what the speaker will say from the speaker's perspective, predicting “wicked” or “the cup,” even though “wicked” might connote a negative valance to the addressee (rather than the intended positive valance), and even though “the cup” would be ambiguous from the addressee's perspective. Therefore, even though the prediction is executed using the addressee's production system, the entire prediction process would have to be tailored to the addressee's beliefs about the speaker's context.
In our view, the process would be no different on an association view in which addressees predict based on previous perceptual experience. P&G propose that association occurs when interlocutors are dissimilar or the production system is not engaged (e.g., in reading). However, even when unfamiliar interlocutors have different perspectives, overwhelming evidence now suggests that listeners do not progress egocentrically, but instead take into account information about their partner's context (e.g., Hanna et al. Reference Hanna, Tanenhaus and Trueswell2003; Heller et al. Reference Heller, Grodner and Tanenhaus2008). Similarly, as in live conversation, readers make rapid predictions when reading (Federmeier & Kutas Reference Federmeier and Kutas1999) and tailor referential interpretation based on representations of the number of entities in the discourse context (Greene et al. Reference Greene, McKoon and Ratcliff1992; Nieuwland et al. Reference Nieuwland, Otten and Van Berkum2007). Hence, it would seem that prediction-by-association would have to be tailored to the particular context of language use, just as in simulation. What advantage, then, is gained by positing a second mechanism to prediction? P&G suggest association might not afford rapid turn-taking, however this seems less of an argument to posit this mechanism than an argument against it, given that turn-taking is, in fact, rapid. P&G also suggest that individuals might choose to use either association, simulation, or a combination of the two; however, it is unclear how these decisions would be made, and how the outputs of these mechanisms would be integrated during real-time processing.
In conclusion, we applaud P&G's emphasis on the way production and comprehension are interwoven in natural communication. However, in emphasizing the remarkable skill needed to produce, for example, a split turn, the authors overlook potential redundancy in the dual-mechanism proposal. Indeed, the evidence suggests that in a large variety of circumstances, interlocutors integrate context and common ground into processing predictions. The accuracy, speed, and type of prediction seem to be determined largely by factors such as the quality of the listener's estimation of the speaker's context, and whether attending to one's partner's context is relevant to the communicative goals (Yoon et al. Reference Yoon, Koh and Brown-Schmidt2012). Hence, the determining factor in the quality of prediction should be seen as context-modeling, rather than a decision to use one mechanism in a hypothesized processing architecture. Understanding the mechanisms that determine the context of conversation, and the degree to which the contexts of the speaker and listener are coordinated, then, would seem to be a central goal for understanding dialogic processes.
We agree with Pickering & Garrod's (P&G's) claim that models of language use must take into account the fact that production and comprehension processes are interwoven in time and interconnected as in the case of split turns. Indeed, the most basic form of language use is arguably conversation, in which interlocutors act both as producers and addressees, and each type of act involves elements of the other.
The importance of examining dialogic processes has become apparent following a surge of interest in studying language in natural settings (Pickering & Garrod Reference Pickering and Garrod2004; Trueswell & Tanenhaus Reference Trueswell and Tanenhaus2005). This interest has generated new, significant findings in conversation including development of techniques for independently quantifying and predicting the degree of coordination in conversation (Richardson et al. Reference Richardson, Dale and Kirkham2007), as well as evidence that coordinated contextual representations facilitate use of potentially ambiguous referential expressions (Brown-Schmidt & Tanenhaus Reference Brown-Schmidt and Tanenhaus2008). Similarly, experiments in noninteractive settings increasingly focus on dialogue-relevant questions such as how representations of others' mental states guide processing (Ferguson et al. Reference Ferguson, Scheepers and Sanford2010).
A central feature of language use is that it is produced and understood with respect to a particular context that constrains both what we say and how we say it. In conversation, the basic context is often assumed to be the interlocutors' common ground (Clark Reference Clark1996) with conversational efficiency increasing as common ground grows (Wilkes-Gibbs & Clark Reference Wilkes-Gibbs and Clark1992). According to P&G's proposal, context plays a key role in circumscribing prediction and, as a result, conversational efficiency.
According to P&G, listeners predict upcoming utterances using one of two mechanisms, simulation and association. Listeners use the simulation mechanism with familiar partners, and the association mechanism when they perceive themselves as being different from their partner, as in adult-child conversations, and in reading, where the addressee does not typically speak. In what follows, we argue that with both familiar and unfamiliar partners, and in talking and reading alike, language users make sophisticated predictions about future language use, based on available contextual information.
P&G argue that an addressee using the simulation mechanism will predict an utterance using an inverse model and contextual information – a prediction about what the speaker would say. P&G do not specify the details of this contextual information. However, to predict what the speaker would say requires assessing the speaker's context, where context must be broadly defined to include both local perceptual information as well as historical information about the person's dialect and past experience. After all, depending on a person's age and regional dialect, a given semantic meaning might be expressed as great, rad, or wicked. Similarly, in cases where a potential alternative referent is occluded from the speaker's but not the addressee's view (e.g., the speaker sees one cup whereas the addressee sees two), the context would predict different referential expressions depending on which perspective was used. We read P&G as saying that in each of these cases, on the simulation mechanism, the addressee would predict what the speaker will say from the speaker's perspective, predicting “wicked” or “the cup,” even though “wicked” might connote a negative valance to the addressee (rather than the intended positive valance), and even though “the cup” would be ambiguous from the addressee's perspective. Therefore, even though the prediction is executed using the addressee's production system, the entire prediction process would have to be tailored to the addressee's beliefs about the speaker's context.
In our view, the process would be no different on an association view in which addressees predict based on previous perceptual experience. P&G propose that association occurs when interlocutors are dissimilar or the production system is not engaged (e.g., in reading). However, even when unfamiliar interlocutors have different perspectives, overwhelming evidence now suggests that listeners do not progress egocentrically, but instead take into account information about their partner's context (e.g., Hanna et al. Reference Hanna, Tanenhaus and Trueswell2003; Heller et al. Reference Heller, Grodner and Tanenhaus2008). Similarly, as in live conversation, readers make rapid predictions when reading (Federmeier & Kutas Reference Federmeier and Kutas1999) and tailor referential interpretation based on representations of the number of entities in the discourse context (Greene et al. Reference Greene, McKoon and Ratcliff1992; Nieuwland et al. Reference Nieuwland, Otten and Van Berkum2007). Hence, it would seem that prediction-by-association would have to be tailored to the particular context of language use, just as in simulation. What advantage, then, is gained by positing a second mechanism to prediction? P&G suggest association might not afford rapid turn-taking, however this seems less of an argument to posit this mechanism than an argument against it, given that turn-taking is, in fact, rapid. P&G also suggest that individuals might choose to use either association, simulation, or a combination of the two; however, it is unclear how these decisions would be made, and how the outputs of these mechanisms would be integrated during real-time processing.
In conclusion, we applaud P&G's emphasis on the way production and comprehension are interwoven in natural communication. However, in emphasizing the remarkable skill needed to produce, for example, a split turn, the authors overlook potential redundancy in the dual-mechanism proposal. Indeed, the evidence suggests that in a large variety of circumstances, interlocutors integrate context and common ground into processing predictions. The accuracy, speed, and type of prediction seem to be determined largely by factors such as the quality of the listener's estimation of the speaker's context, and whether attending to one's partner's context is relevant to the communicative goals (Yoon et al. Reference Yoon, Koh and Brown-Schmidt2012). Hence, the determining factor in the quality of prediction should be seen as context-modeling, rather than a decision to use one mechanism in a hypothesized processing architecture. Understanding the mechanisms that determine the context of conversation, and the degree to which the contexts of the speaker and listener are coordinated, then, would seem to be a central goal for understanding dialogic processes.
ACKNOWLEDGMENTS
Preparation of this article was partially supported by NSF grant no. BCS 10-19161 to S. Brown-Schmidt. Thank you to Jennifer Roche and Kara Federmeier for helpful discussions.