How can we determine the mental representation of linguistic expressions in relation to the time course from intention to articulation and vice versa (Levelt Reference Levelt1989)? A new experimental technique to tap into this process like Branigan & Pickering's (B&P's) contribution is very welcome. Their review of priming experiments shows that expressions with a particular linguistic structure can facilitate the use of other expressions with a certain structural similarity. From this, they draw strong conclusions. Their interpretation is not compelling, however, and occasionally reflects a misanalysis (e.g., the Mandarin topicalization in section 2.4 shows only that an (A'-bound) empty object is visible for priming). We suggest an alternative interpretation of their findings.
Under mainstream generative accounts, B&P argue, passives involve movement of the underlying object to subject position (leaving an NP trace/copy), whereas intransitive (active) locatives do not. Hence, the two sentence types involve different representations. B&P, however, report experiments in which intransitive locatives prime passives. The mainstream account is – they conclude – incompatible with this result.
They also discuss the unergative-unaccusative contrast, which is captured standardly by assuming that the unaccusative argument is first inserted in the DO position and next moved to the canonical subject position. Unergatives don't exhibit such movement. B&P show that, nevertheless, intransitive sentences with unergatives and unaccusatives prime each other. Hence, B&P argue, their difference is not syntactically represented. Syntactic representations, then, must contain much less detail than generative approaches assume: There is no syntactic movement, and syntactic representations do not contain copies/traces.
However, B&P mistakenly infer that anything you cannot see with structural priming is “inaccessible” (not used) in processing (sect. 1.1). As is well known, all experimental techniques aren't sensitive to the same processes. If a property established by one technique is not observed with another technique, it is a fallacy to conclude more than that there is a discrepancy to be explained. Crucially, there is abundant evidence that certain properties that, according to B&P, are not visible for priming are, in fact, visible to the processor.
For example, B&P's claim that the contrast between unaccusatives and unergatives is purely semantic and not syntactically encoded is untenable. First of all, the original tests from Perlmutter (Reference Perlmutter, Jaeger, Woodbury, Ackerman, Chiarello, Gensler, Kingston, Sweetser, Thompson and Whistler1978) and Burzio (Reference Burzio1981) show that, unlike the subject of unergatives, the subject of unaccusatives shares syntactic properties with direct objects. Second, these verb types display a difference in processing that is independent of semantic roles (e.g., Agnew et al. Reference Agnew, van de Koot, McGettigan and Scott2014; Koring et al. Reference Koring, Mak and Reuland2012). It follows from a difference in the structural representation, which, consequently, must be visible to the processor, contra B&P.
A second misconception concerns their claim that structural priming reveals the exact nature of syntactic representations. As B&P point out themselves, priming displays similarities in representation of a pair of sentences A and B relative to the pair A and C (pp.19–20). As such, this measure cannot tell us directly what the representation of a sentence looks like. A priming effect can tell us at most that particular sentences share some aspects of their representation, but this does not entail that their representations can be identified.
B&P's appeal to parsimony in their argumentation also fails due to inaccuracies in their exposition (including references to obsolete concepts like Deep Structure). Moreover, the absence of explicit mapping rules between syntax and semantics makes their preferred alternative impossible to assess. In generative theory (see Chomsky Reference Chomsky1986; Reference Chomsky1995; Reference Chomsky and Kenstowicz2001; also Reference Chomsky1955/1975), the role of grammar is not so much to characterize what is grammatical as opposed to ungrammatical, but to characterize the relation between forms and their interpretations. Due to the phase-based organization of derivations, B&P's reference to levels misses the point. Properties reflecting steps in this derivational process are accessible to the processor, as shown by a variety of experimental techniques currently employed in addition to grammaticality judgment tasks (which B&P fail to acknowledge) (e.g., Bever & Sanz Reference Bever and Sanz1997; Brennan & Pylkkänen Reference Brennan and Pylkkänen2016; Crain & Thornton Reference Crain and Thornton1998; Friedmann et al. Reference Friedmann, Taranto and Swinney2008; Koornneef et al. Reference Koornneef, Avrutin, Wijnen, Reuland and Runner2011; Sprouse et al. Reference Sprouse, Caponigro, Greco and Cecchetto2016).
Unlike what B&P presume, (Narrow) Syntax is independently characterized, namely as involving operations subject to restrictions (e.g., locality constraints) that are independent of intended meaning. Consider resumptive pronouns in wh-questions. The formation of wh-questions is subject to locality conditions. Interestingly, wh-questions that violate a locality condition can be “saved” by using a resumptive pronoun. The resumptive pronoun does not contribute to the meaning but makes an otherwise ungrammatical dependency licit. This shows that the interpretation itself is not blocked, but a particular syntactic derivation to realize that interpretation (for a similar contrast in binding dependencies, see e.g. Reuland Reference Reuland2011a; Reference Reuland and Boeckx2011b; Koornneef & Reuland Reference Koornneef and Reuland2016). Therefore, B&P's argument that there is no level of detailed syntactic representation because the priming tool does not track that level is misguided.
Yet, we share B&P's concern “[to identify] which aspect of structure that priming taps into” (sect. 1.4, para. 7). We suggest that the method of structural priming tracks no more, but also no less, than a particular aspect of detailed linguistic representations – namely, what is visible to the external systems. Phase theory helps us identify this aspect. Phase theory hypothesizes that, once the derivation of a relevant chunk – a propositional structure, a DP/PP – is complete, it is handed over to the realization and interpretation systems. Its internal structure – copies/traces – becomes inaccessible at that point. Hence, at this handover point, what is accessible in unergative and unaccusative structures will be quite similar, yielding the priming data unsurprising. The same applies to passives. What is visible of their internal structure will lack detail at the handover point, making them sufficiently similar to locatives for priming. Finally, given that scope marking is structurally represented, and the scope marker is external to the core proposition, the latter's internal structure, but not the scope marker, will have become inaccessible at the handover point. This reinterpretation in terms of phases provides a straightforward account of B&P's findings. In short, phase theory can help understand what structural priming shows.
You have
Access
How can we determine the mental representation of linguistic expressions in relation to the time course from intention to articulation and vice versa (Levelt Reference Levelt1989)? A new experimental technique to tap into this process like Branigan & Pickering's (B&P's) contribution is very welcome. Their review of priming experiments shows that expressions with a particular linguistic structure can facilitate the use of other expressions with a certain structural similarity. From this, they draw strong conclusions. Their interpretation is not compelling, however, and occasionally reflects a misanalysis (e.g., the Mandarin topicalization in section 2.4 shows only that an (A'-bound) empty object is visible for priming). We suggest an alternative interpretation of their findings.
Under mainstream generative accounts, B&P argue, passives involve movement of the underlying object to subject position (leaving an NP trace/copy), whereas intransitive (active) locatives do not. Hence, the two sentence types involve different representations. B&P, however, report experiments in which intransitive locatives prime passives. The mainstream account is – they conclude – incompatible with this result.
They also discuss the unergative-unaccusative contrast, which is captured standardly by assuming that the unaccusative argument is first inserted in the DO position and next moved to the canonical subject position. Unergatives don't exhibit such movement. B&P show that, nevertheless, intransitive sentences with unergatives and unaccusatives prime each other. Hence, B&P argue, their difference is not syntactically represented. Syntactic representations, then, must contain much less detail than generative approaches assume: There is no syntactic movement, and syntactic representations do not contain copies/traces.
However, B&P mistakenly infer that anything you cannot see with structural priming is “inaccessible” (not used) in processing (sect. 1.1). As is well known, all experimental techniques aren't sensitive to the same processes. If a property established by one technique is not observed with another technique, it is a fallacy to conclude more than that there is a discrepancy to be explained. Crucially, there is abundant evidence that certain properties that, according to B&P, are not visible for priming are, in fact, visible to the processor.
For example, B&P's claim that the contrast between unaccusatives and unergatives is purely semantic and not syntactically encoded is untenable. First of all, the original tests from Perlmutter (Reference Perlmutter, Jaeger, Woodbury, Ackerman, Chiarello, Gensler, Kingston, Sweetser, Thompson and Whistler1978) and Burzio (Reference Burzio1981) show that, unlike the subject of unergatives, the subject of unaccusatives shares syntactic properties with direct objects. Second, these verb types display a difference in processing that is independent of semantic roles (e.g., Agnew et al. Reference Agnew, van de Koot, McGettigan and Scott2014; Koring et al. Reference Koring, Mak and Reuland2012). It follows from a difference in the structural representation, which, consequently, must be visible to the processor, contra B&P.
A second misconception concerns their claim that structural priming reveals the exact nature of syntactic representations. As B&P point out themselves, priming displays similarities in representation of a pair of sentences A and B relative to the pair A and C (pp.19–20). As such, this measure cannot tell us directly what the representation of a sentence looks like. A priming effect can tell us at most that particular sentences share some aspects of their representation, but this does not entail that their representations can be identified.
B&P's appeal to parsimony in their argumentation also fails due to inaccuracies in their exposition (including references to obsolete concepts like Deep Structure). Moreover, the absence of explicit mapping rules between syntax and semantics makes their preferred alternative impossible to assess. In generative theory (see Chomsky Reference Chomsky1986; Reference Chomsky1995; Reference Chomsky and Kenstowicz2001; also Reference Chomsky1955/1975), the role of grammar is not so much to characterize what is grammatical as opposed to ungrammatical, but to characterize the relation between forms and their interpretations. Due to the phase-based organization of derivations, B&P's reference to levels misses the point. Properties reflecting steps in this derivational process are accessible to the processor, as shown by a variety of experimental techniques currently employed in addition to grammaticality judgment tasks (which B&P fail to acknowledge) (e.g., Bever & Sanz Reference Bever and Sanz1997; Brennan & Pylkkänen Reference Brennan and Pylkkänen2016; Crain & Thornton Reference Crain and Thornton1998; Friedmann et al. Reference Friedmann, Taranto and Swinney2008; Koornneef et al. Reference Koornneef, Avrutin, Wijnen, Reuland and Runner2011; Sprouse et al. Reference Sprouse, Caponigro, Greco and Cecchetto2016).
Unlike what B&P presume, (Narrow) Syntax is independently characterized, namely as involving operations subject to restrictions (e.g., locality constraints) that are independent of intended meaning. Consider resumptive pronouns in wh-questions. The formation of wh-questions is subject to locality conditions. Interestingly, wh-questions that violate a locality condition can be “saved” by using a resumptive pronoun. The resumptive pronoun does not contribute to the meaning but makes an otherwise ungrammatical dependency licit. This shows that the interpretation itself is not blocked, but a particular syntactic derivation to realize that interpretation (for a similar contrast in binding dependencies, see e.g. Reuland Reference Reuland2011a; Reference Reuland and Boeckx2011b; Koornneef & Reuland Reference Koornneef and Reuland2016). Therefore, B&P's argument that there is no level of detailed syntactic representation because the priming tool does not track that level is misguided.
Yet, we share B&P's concern “[to identify] which aspect of structure that priming taps into” (sect. 1.4, para. 7). We suggest that the method of structural priming tracks no more, but also no less, than a particular aspect of detailed linguistic representations – namely, what is visible to the external systems. Phase theory helps us identify this aspect. Phase theory hypothesizes that, once the derivation of a relevant chunk – a propositional structure, a DP/PP – is complete, it is handed over to the realization and interpretation systems. Its internal structure – copies/traces – becomes inaccessible at that point. Hence, at this handover point, what is accessible in unergative and unaccusative structures will be quite similar, yielding the priming data unsurprising. The same applies to passives. What is visible of their internal structure will lack detail at the handover point, making them sufficiently similar to locatives for priming. Finally, given that scope marking is structurally represented, and the scope marker is external to the core proposition, the latter's internal structure, but not the scope marker, will have become inaccessible at the handover point. This reinterpretation in terms of phases provides a straightforward account of B&P's findings. In short, phase theory can help understand what structural priming shows.
ACKNOWLEDGMENTS
We are very grateful to Noam Chomsky, Martin Everaert, Arnout Koornneef, Pim Mak, and Iris Mulders for their very helpful comments on an earlier draft. Any errors are ours.