I agree with the authors that the way language is represented in the human mind is a central question for the language sciences; that the question must be studied empirically; that obtaining acceptability judgements from experts or laypersons is not sufficient to do so; and that important evidence about the nature of linguistic representations can be gleaned from structural priming (SP) experiments. The fruitfulness of the paradigm is amply illustrated in the theoretical part of the target paper and by the papers in Dell and Ferreira (Reference Dell and Ferreira2016).
I am not convinced, however, that the paradigm is as versatile as the authors suggest. Designing and running SP experiments is not trivial. As Mahowald et al. (Reference Mahowald, James, Futrell and Gibson2016b) show in a meta-analysis, SP effects are usually small to moderate, and for adequately powered designs, many items and/or participants are needed. This limits the usefulness of the paradigm for studies involving children and patients. Moreover, for secure interpretation of SP effects, many conceptual and linguistic variables (frequencies of words and word combinations, plausibility, etc.) must be taken into account; for production experiments, suitable ways of reliably eliciting the target utterances must be found.
Of course, similar considerations hold for other paradigms. One issue, however, pertains specifically to SP – in particular, to the canonical variant of the paradigm in which the speakers' choices of grammatical structures are recorded. As the authors discuss, the design of SP experiments requires the existence of pairs of roughly equivalent structures (e.g., active and passive) for participants to choose between. Therefore, it is challenging to use SP in research on properties of grammatical representations, such as subject-verb agreement or fixed word order (e.g., in English polar questions), for which no alternatives exist in the language. Thus, the authors' statement that “priming can be used similarly to investigate the representation of any aspect of linguistic structure” (sect. 1.4, para. 4) seems overly optimistic.
I am also not convinced that, as B&P propose in section 1.4 of the target paper, “priming effects arguably implicate a direct relationship between representation and behavior” (sect. 1.4, para 2; see also sect. 1.5, para. 1: “It [structural priming] provides evidence that is directly informative about mental representations”). They propose that comprehension-to-production priming reflects on shared representations rather than processing components specific to production or comprehension; they also argue that SP effects differ from judgements, chronometric, or neurobiological data collected in other paradigms by being uncontaminated by processing influences: “It is hard to see how the explanation of priming could depend on processing assumptions” (sect. 1.4, para. 9). The authors acknowledge that priming may occur for reasons other than similarity of representations, but they believe that such effects should be easy to identify through careful experimentation.
It is not clear to me why comprehension-to-production priming necessarily reflects on shared representations rather than shared processes, and why the authors believe that it is easy to separate effects of the similarity of representations from other causes of priming. They point out that priming effects occur “without awareness or explicit recall of the prime stimulus and are generally believed to be automatic and resource free” (sect. 1.4, para. 2). However, processes that occur without awareness and that are largely automatic are still processes. Most important, these processes change the speaker's language use: After presentation of a prime, the primed structure is more readily available than it was before. Thus, SP effects do not only reflect on the degree of similarity between linguistic representations or processes, but also on their malleability. This is why SP often is seen as a form of implicit learning (e.g. Bock & Griffin Reference Bock and Griffin2000; Chang et al. Reference Chang, Dell and Bock2006).
In addition to changing the availability of known structural alternatives, SP can introduce new constructions into a person's repertoire. For instance, in a reading study, Fraundorf and Jaeger (Reference Fraundorf and Jaeger2016) showed priming for sentences such as The car needs cleaned, which featured a structure (need + past participle) that was not part of the participants' dialect. This effect generalized to structurally similar sentences featuring will + past participle, as in The copier will recycled. Interestingly, facilitation for will sentences after exposure to need sentences occurred only in participants who were not familiar with the need + past participle construction before the experiment but not in participants whose dialect permitted this construction. Fraundorf and Jaeger speculated that this pattern arose because only the former group expected encountering unfamiliar structures. Whatever the merits of this specific account, the results illustrate, first, that SP can alter a person's linguistic repertoire, and, second, that such changes are subject to multiple influences, including the person's expectations about the utterances likely to occur in the current context. This latter conclusion is supported by several other recent studies (e.g., Myslín & Levy Reference Myslín and Levy2016). Fraundorf and Jaeger reported a comprehension study. Maybe speakers' choices in production experiments are less readily affected by other variables than prime-target similarity. It seems highly unlikely, however, that they are entirely impervious to such influences, or, as B&P suggest, that it is trivial to separate them from structural similarity effects.
In sum, the claim that SP offers an unobstructed route to linguistic representations seems incorrect to me. SP shows how grammatical choices change with experience. Choices are behaviour, and like any other type of complex behaviour, they are based on stored knowledge and cognitive processes using this knowledge. The authors note that acceptability judgements suffer from “source ambiguity” – uncertainty about the origins of observed effects. Exactly the same holds for SP and, in fact, any other psycholinguistic
I agree with the authors that the way language is represented in the human mind is a central question for the language sciences; that the question must be studied empirically; that obtaining acceptability judgements from experts or laypersons is not sufficient to do so; and that important evidence about the nature of linguistic representations can be gleaned from structural priming (SP) experiments. The fruitfulness of the paradigm is amply illustrated in the theoretical part of the target paper and by the papers in Dell and Ferreira (Reference Dell and Ferreira2016).
I am not convinced, however, that the paradigm is as versatile as the authors suggest. Designing and running SP experiments is not trivial. As Mahowald et al. (Reference Mahowald, James, Futrell and Gibson2016b) show in a meta-analysis, SP effects are usually small to moderate, and for adequately powered designs, many items and/or participants are needed. This limits the usefulness of the paradigm for studies involving children and patients. Moreover, for secure interpretation of SP effects, many conceptual and linguistic variables (frequencies of words and word combinations, plausibility, etc.) must be taken into account; for production experiments, suitable ways of reliably eliciting the target utterances must be found.
Of course, similar considerations hold for other paradigms. One issue, however, pertains specifically to SP – in particular, to the canonical variant of the paradigm in which the speakers' choices of grammatical structures are recorded. As the authors discuss, the design of SP experiments requires the existence of pairs of roughly equivalent structures (e.g., active and passive) for participants to choose between. Therefore, it is challenging to use SP in research on properties of grammatical representations, such as subject-verb agreement or fixed word order (e.g., in English polar questions), for which no alternatives exist in the language. Thus, the authors' statement that “priming can be used similarly to investigate the representation of any aspect of linguistic structure” (sect. 1.4, para. 4) seems overly optimistic.
I am also not convinced that, as B&P propose in section 1.4 of the target paper, “priming effects arguably implicate a direct relationship between representation and behavior” (sect. 1.4, para 2; see also sect. 1.5, para. 1: “It [structural priming] provides evidence that is directly informative about mental representations”). They propose that comprehension-to-production priming reflects on shared representations rather than processing components specific to production or comprehension; they also argue that SP effects differ from judgements, chronometric, or neurobiological data collected in other paradigms by being uncontaminated by processing influences: “It is hard to see how the explanation of priming could depend on processing assumptions” (sect. 1.4, para. 9). The authors acknowledge that priming may occur for reasons other than similarity of representations, but they believe that such effects should be easy to identify through careful experimentation.
It is not clear to me why comprehension-to-production priming necessarily reflects on shared representations rather than shared processes, and why the authors believe that it is easy to separate effects of the similarity of representations from other causes of priming. They point out that priming effects occur “without awareness or explicit recall of the prime stimulus and are generally believed to be automatic and resource free” (sect. 1.4, para. 2). However, processes that occur without awareness and that are largely automatic are still processes. Most important, these processes change the speaker's language use: After presentation of a prime, the primed structure is more readily available than it was before. Thus, SP effects do not only reflect on the degree of similarity between linguistic representations or processes, but also on their malleability. This is why SP often is seen as a form of implicit learning (e.g. Bock & Griffin Reference Bock and Griffin2000; Chang et al. Reference Chang, Dell and Bock2006).
In addition to changing the availability of known structural alternatives, SP can introduce new constructions into a person's repertoire. For instance, in a reading study, Fraundorf and Jaeger (Reference Fraundorf and Jaeger2016) showed priming for sentences such as The car needs cleaned, which featured a structure (need + past participle) that was not part of the participants' dialect. This effect generalized to structurally similar sentences featuring will + past participle, as in The copier will recycled. Interestingly, facilitation for will sentences after exposure to need sentences occurred only in participants who were not familiar with the need + past participle construction before the experiment but not in participants whose dialect permitted this construction. Fraundorf and Jaeger speculated that this pattern arose because only the former group expected encountering unfamiliar structures. Whatever the merits of this specific account, the results illustrate, first, that SP can alter a person's linguistic repertoire, and, second, that such changes are subject to multiple influences, including the person's expectations about the utterances likely to occur in the current context. This latter conclusion is supported by several other recent studies (e.g., Myslín & Levy Reference Myslín and Levy2016). Fraundorf and Jaeger reported a comprehension study. Maybe speakers' choices in production experiments are less readily affected by other variables than prime-target similarity. It seems highly unlikely, however, that they are entirely impervious to such influences, or, as B&P suggest, that it is trivial to separate them from structural similarity effects.
In sum, the claim that SP offers an unobstructed route to linguistic representations seems incorrect to me. SP shows how grammatical choices change with experience. Choices are behaviour, and like any other type of complex behaviour, they are based on stored knowledge and cognitive processes using this knowledge. The authors note that acceptability judgements suffer from “source ambiguity” – uncertainty about the origins of observed effects. Exactly the same holds for SP and, in fact, any other psycholinguistic