Branigan & Pickering (B&P) argue that evidence from structural priming informs our understanding of abstract linguistic representations in ways that traditional acceptability judgments do not, thus providing a valuable tool for developing psychologically plausible theories of syntax and semantics. I agree with them on this point and believe their approach can shed light on tricky cases of grammaticalization in progress, as suggested below. However, I maintain that evidence from structural priming can be ambiguous with respect to the influence of syntactic structure alone versus syntax-semantics mappings and, therefore, cannot distinguish by itself between competing syntactic analyses.
Examining data from dozens of published studies that have used structural priming to investigate language processing, B&P propose a theoretical approach similar to Culicover and Jackendoff's (Reference Culicover and Jackendoff2005) Simpler Syntax. According to B&P's approach, syntax consists of a “shallow” constituent structure without any movement transformations and minimal null constituents; thematic roles, event structure, and quantificational information are included only in semantics. Supporting this view, they cite evidence showing that speakers are sensitive to shallow syntax even when the semantic argument structure differs between prime and target (Bock & Loebell Reference Bock and Loebell1990; Flett Reference Flett2006; Wittenberg Reference Wittenberg2014). I suggest that B&P's approach can be fruitfully applied also to cases in which nouns, verbs, or adjectives have undergone partial grammaticalization. Such cases are notoriously challenging for synchronic theories of syntactic representation because the items in question show mixed properties of lexical and functional categories (Denison Reference Denison, Traugott and Trousdale2010). Francis and Yuasa (Reference Francis and Yuasa2008) argued, based on evidence from English, Japanese, and Cantonese, that at least some such cases involve semantic change in the absence of syntactic reanalysis – a phenomenon they captured synchronically using a parallel-architecture representation (Culicover & Jackendoff Reference Culicover and Jackendoff2005; Sadock Reference Sadock1991). For example, English quantificational nouns (e.g., lot, bunch, ton) display mixed properties of collective nouns (e.g., bundle), and quantifiers (e.g., many), with the quantifier-like properties due entirely to semantics. Thus, quantificational NPs (e.g., a lot of sticks) share a syntactic representation with collective NPs (e.g., a bundle of sticks), despite differences in meaning. In both types of NP, the first noun (lot, bundle) acts as the syntactic head of the phrase. This analysis relied on attributing some patterns of acceptability judgments to syntax and others, to semantics. Following B&P's proposal, one could test whether collective NPs would prime the production of quantificational NPs when speakers are asked to describe a set of objects in terms of quantity. Such a priming effect would support Francis and Yuasa's proposal that quantificational nouns really do act as head nouns in syntax, while the absence of any priming effect would suggest that quantificational NPs and collective NPs differ syntactically. More generally, at least for cases in which the source construction continues to exist in the language alongside the grammaticalized form, priming tasks could help determine whether a lexical item or construction that has undergone semantic change also has undergone syntactic reanalysis.
It is less clear, however, how to interpret situations in which priming effects appear in different degrees for different types of primes. B&P cite studies showing an enhanced priming effect when the prime and target share not only the same shallow syntactic representation, but also the same abstract syntax-semantics mapping (Griffin & Weinstein-Tull Reference Griffin and Weinstein-Tull2003; Raffray et al. Reference Raffray, Pickering, Cai and Branigan2014). Raffray et al. (Reference Raffray, Pickering, Cai and Branigan2014) found that sentences with a coerced (missing) predicate (e.g., The celebrity began the champagne) primed target responses with a coerced predicate more effectively than did syntactically similar sentences with an event NP (e.g., The celebrity began the speech). However, these two types of NP-V-NP sentences were alike in failing to prime NP-V-VP responses (e.g., The celebrity began drinking the champagne). B&P take the latter fact to mean that the missing predicate of a coerced sentence is not represented in the syntax. But what, then, do we make of the fact that coerced predicate sentences primed coerced predicate responses more strongly than event NP sentences did? Raffray et al. (Reference Raffray, Pickering, Cai and Branigan2014, p. 97) propose that speakers were sensitive to particular syntax-semantics mappings, in addition to being sensitive to shallow syntax. It seems, however, that one could plausibly interpret the different degrees of priming to signal some subtle difference in the syntactic representations.
Similarly, Griffin and Weinstein-Tull (Reference Griffin and Weinstein-Tull2003) found that object-raising infinitives (e.g., John believed Mary to be nice) primed object-raising responses (as opposed to finite paraphrases) more effectively than object-control infinitives did (e.g., John persuaded Mary to be nice). Griffin and Weinstein-Tull (Reference Griffin and Weinstein-Tull2003, p. 549) interpreted these findings to mean that language users are sensitive to the similarity of abstract syntax-semantics mappings between prime and target, because object-control infinitives have an additional argument role. As an alternative to this explanation, B&P suggest that perhaps the additional argument role in object-control sentences is associated with a distinct syntactic representation, meaning that reference to syntax-semantics mapping is not necessary to explain the results (note 6 in the target article). While Griffin and Weinstein-Tull's explanation is compatible with parallel-architecture theories such as Simpler Syntax (Culicover & Jackendoff Reference Culicover and Jackendoff2005) and Construction Grammar (Goldberg Reference Goldberg2006) – which do not permit null constituents in the syntax – B&P's alternative requires a null constituent in the syntax. Therefore, B&P's explanation is more compatible with the standard generative account of object control, in which the infinitive subject is a null pronoun (PRO) (Chomsky Reference Chomsky1981). As B&P acknowledge, the evidence does not distinguish between these two possible syntactic representations.
Evidence from priming is useful for showing speakers' sensitivity to structural similarity. As such, it can indicate which elements must be included in syntactic representations (e.g., syntactic categories, constituent ordering) and whether historically related constructions continue to share a constituent structure. It cannot tell us, however, whether differences in structure, as reflected in differences in degrees of priming, are due to differences in syntactic representation alone or to syntax-semantics mappings. While structural priming provides valuable evidence for linguistic theory building, the abstract representational basis for any particular facilitation effect remains subject to interpretation. A psychologically plausible theory of syntactic and semantic representation, therefore, must take into account many different types of evidence, each of which has its own advantages and limitations.
Branigan & Pickering (B&P) argue that evidence from structural priming informs our understanding of abstract linguistic representations in ways that traditional acceptability judgments do not, thus providing a valuable tool for developing psychologically plausible theories of syntax and semantics. I agree with them on this point and believe their approach can shed light on tricky cases of grammaticalization in progress, as suggested below. However, I maintain that evidence from structural priming can be ambiguous with respect to the influence of syntactic structure alone versus syntax-semantics mappings and, therefore, cannot distinguish by itself between competing syntactic analyses.
Examining data from dozens of published studies that have used structural priming to investigate language processing, B&P propose a theoretical approach similar to Culicover and Jackendoff's (Reference Culicover and Jackendoff2005) Simpler Syntax. According to B&P's approach, syntax consists of a “shallow” constituent structure without any movement transformations and minimal null constituents; thematic roles, event structure, and quantificational information are included only in semantics. Supporting this view, they cite evidence showing that speakers are sensitive to shallow syntax even when the semantic argument structure differs between prime and target (Bock & Loebell Reference Bock and Loebell1990; Flett Reference Flett2006; Wittenberg Reference Wittenberg2014). I suggest that B&P's approach can be fruitfully applied also to cases in which nouns, verbs, or adjectives have undergone partial grammaticalization. Such cases are notoriously challenging for synchronic theories of syntactic representation because the items in question show mixed properties of lexical and functional categories (Denison Reference Denison, Traugott and Trousdale2010). Francis and Yuasa (Reference Francis and Yuasa2008) argued, based on evidence from English, Japanese, and Cantonese, that at least some such cases involve semantic change in the absence of syntactic reanalysis – a phenomenon they captured synchronically using a parallel-architecture representation (Culicover & Jackendoff Reference Culicover and Jackendoff2005; Sadock Reference Sadock1991). For example, English quantificational nouns (e.g., lot, bunch, ton) display mixed properties of collective nouns (e.g., bundle), and quantifiers (e.g., many), with the quantifier-like properties due entirely to semantics. Thus, quantificational NPs (e.g., a lot of sticks) share a syntactic representation with collective NPs (e.g., a bundle of sticks), despite differences in meaning. In both types of NP, the first noun (lot, bundle) acts as the syntactic head of the phrase. This analysis relied on attributing some patterns of acceptability judgments to syntax and others, to semantics. Following B&P's proposal, one could test whether collective NPs would prime the production of quantificational NPs when speakers are asked to describe a set of objects in terms of quantity. Such a priming effect would support Francis and Yuasa's proposal that quantificational nouns really do act as head nouns in syntax, while the absence of any priming effect would suggest that quantificational NPs and collective NPs differ syntactically. More generally, at least for cases in which the source construction continues to exist in the language alongside the grammaticalized form, priming tasks could help determine whether a lexical item or construction that has undergone semantic change also has undergone syntactic reanalysis.
It is less clear, however, how to interpret situations in which priming effects appear in different degrees for different types of primes. B&P cite studies showing an enhanced priming effect when the prime and target share not only the same shallow syntactic representation, but also the same abstract syntax-semantics mapping (Griffin & Weinstein-Tull Reference Griffin and Weinstein-Tull2003; Raffray et al. Reference Raffray, Pickering, Cai and Branigan2014). Raffray et al. (Reference Raffray, Pickering, Cai and Branigan2014) found that sentences with a coerced (missing) predicate (e.g., The celebrity began the champagne) primed target responses with a coerced predicate more effectively than did syntactically similar sentences with an event NP (e.g., The celebrity began the speech). However, these two types of NP-V-NP sentences were alike in failing to prime NP-V-VP responses (e.g., The celebrity began drinking the champagne). B&P take the latter fact to mean that the missing predicate of a coerced sentence is not represented in the syntax. But what, then, do we make of the fact that coerced predicate sentences primed coerced predicate responses more strongly than event NP sentences did? Raffray et al. (Reference Raffray, Pickering, Cai and Branigan2014, p. 97) propose that speakers were sensitive to particular syntax-semantics mappings, in addition to being sensitive to shallow syntax. It seems, however, that one could plausibly interpret the different degrees of priming to signal some subtle difference in the syntactic representations.
Similarly, Griffin and Weinstein-Tull (Reference Griffin and Weinstein-Tull2003) found that object-raising infinitives (e.g., John believed Mary to be nice) primed object-raising responses (as opposed to finite paraphrases) more effectively than object-control infinitives did (e.g., John persuaded Mary to be nice). Griffin and Weinstein-Tull (Reference Griffin and Weinstein-Tull2003, p. 549) interpreted these findings to mean that language users are sensitive to the similarity of abstract syntax-semantics mappings between prime and target, because object-control infinitives have an additional argument role. As an alternative to this explanation, B&P suggest that perhaps the additional argument role in object-control sentences is associated with a distinct syntactic representation, meaning that reference to syntax-semantics mapping is not necessary to explain the results (note 6 in the target article). While Griffin and Weinstein-Tull's explanation is compatible with parallel-architecture theories such as Simpler Syntax (Culicover & Jackendoff Reference Culicover and Jackendoff2005) and Construction Grammar (Goldberg Reference Goldberg2006) – which do not permit null constituents in the syntax – B&P's alternative requires a null constituent in the syntax. Therefore, B&P's explanation is more compatible with the standard generative account of object control, in which the infinitive subject is a null pronoun (PRO) (Chomsky Reference Chomsky1981). As B&P acknowledge, the evidence does not distinguish between these two possible syntactic representations.
Evidence from priming is useful for showing speakers' sensitivity to structural similarity. As such, it can indicate which elements must be included in syntactic representations (e.g., syntactic categories, constituent ordering) and whether historically related constructions continue to share a constituent structure. It cannot tell us, however, whether differences in structure, as reflected in differences in degrees of priming, are due to differences in syntactic representation alone or to syntax-semantics mappings. While structural priming provides valuable evidence for linguistic theory building, the abstract representational basis for any particular facilitation effect remains subject to interpretation. A psychologically plausible theory of syntactic and semantic representation, therefore, must take into account many different types of evidence, each of which has its own advantages and limitations.