I fully agree with Branigan & Pickering (B&P) that it is high time psychologists, psycholinguists, and linguists join forces in investigating linguistic knowledge. I also agree that priming experiments can contribute to arriving at a better, cognitively more realistic understanding of language. Priming evidence, however, is not immune to theoretical bias. I will illustrate this by approaching the priming effects reported by B&P from the perspective of usage-based cognitive linguistics (UBCL) (e.g., Croft Reference Croft2001; Goldberg Reference Goldberg1995, Reference Goldberg2003, Reference Goldberg2006; Langacker Reference Langacker, Barlow and Kemmer2000, Reference Langacker2008; The Five Graces Group Reference Beckner, Blythe, Bybee, Christiansen, Croft, Ellis, Holland, Ke, Larsen-Freeman, Schoenemann, Ellis and Larsen-Freeman2009; Schmid Reference Schmid2015); that is, I will (re-)interpret them based on the following assumptions:
-
“Language is an integral part of human cognition” (Langacker Reference Langacker1987, p. 12);
-
Linguistic knowledge is highly dynamic and context-adaptive; it emerges from situated usage and is therefore potentially changed by every single usage experience;
-
Rules emerge from generalization processes; they take the form of schematic form–meaning pairs (constructions); that is, they are equivalent in nature and structure to specific lexical items or multi-word units. Consequently, grammar/syntax and lexicon form a continuum;
-
Knowledge of language is organized in a complex network of constructions. This network:
-
• Licenses redundancy, that is, contains both specific linguistic structures and their different degrees of abstractions, as well as both larger (e.g., sentence-level) and smaller constructions (e.g., single lexical items);
-
• Is characterized by multiple (types of) relations that the single constructions (as well as their form- or meaning-sides only) can hold to each other.
These UBCL principles suggest alternative interpretations of priming effects to those proposed by B&P.
Most centrally, they allow for structural priming to be interpreted differently than as evidence in support of the autonomy of syntax. In UBCL terms, the only relevant difference between structural and (more) lexical priming effects is that schematic rather than specific form–meaning pairs are primed. This interpretation allows for all instances of priming (including lexical priming) to be modeled in a unitary fashion: as based on the same mechanisms and as applying to different levels of the same network of linguistic knowledge. Because this network also contains semi-specific elements (e.g., [NP go PP]; see, e.g., Goldberg Reference Goldberg2003), the UBCL view of priming also readily accommodates lexical boost effects. Thus, from a UBCL perspective, the major potential of using priming to investigate linguistic knowledge does not so much lie in differentiating between syntactic and lexical processes and representations, but in identifying which level(s) of the constructional network language users actually and/or habitually access (see also Tomasello Reference Tomasello2006, p. 3).
This construction-based UBCL view also suggests that structural priming effects (just as lexical priming effects; see B&P, sect. 1.4, para. 3) might not occur only when “prime and target involve the same construction” (B&P, note 10), but also when the constructions involved in prime and target processing are (closely) related. The relations driving priming effects can be of different types (e.g., schema–instance, part–whole, similarity, contrast, etc.) and can variably hold between entire constructions or between their form- or meaning-sides only (e.g., Langacker Reference Langacker1987, p. 13). Findings of priming between idioms and formally parallel non-idiomatic utterances (Konopka & Bock Reference Konopka and Bock2009), therefore, do not contradict construction-based UBCL models of language (as claimed by B&P, note 10). The same accounts for priming between sentences in which different thematic roles and/or syntactic functions are formally realized in the same way (e.g., B&P, sect. 2.1, para. 2; Pickering & Ferreira Reference Pickering and Ferreira2008). Instead, from a UBCL perspective, such findings are potentially informative of the internal relational structure of linguistic knowledge and the way it is used in processing.
Priming experiments thus also prove a valuable tool for language research when approached from a UBCL perspective. However, some UBCL assumptions call into question what B&P define as one main benefit of structural priming – namely, that it “provides evidence that is directly informative about mental representation” (sect. 1.5, para. 1).
One of those assumptions is that information is represented redundantly in the language network. It suggests that very different kinds and combinations of constructions can be involved in the processing of what is – formally – the same utterance (e.g., Günther Reference Günther2016, p. 143; Stefanowitsch & Flach Reference Stefanowitsch, Flach and Schmid2016, pp. 105–107); that is, an utterance such as The bottle is on the table might variably be accessed as a single chunk or involve the integration of several more schematic and more specific constructions (e.g., [NP is on NP] + [the bottle] + [the table], etc.). It is thus hardly possible to draw reliable conclusions about which and how many constructions a speaker used when producing or comprehending a particular utterance in a priming experiment.
What complicates matters even further is that, on a usage-based view (e.g., Günther Reference Günther2016), the sets of constructions language users access during utterance production or comprehension are highly likely to differ among individuals and may vary even within individuals across external (situational, social, and linguistic) contextual conditions, as well as across time. Interindividual and cross-contextual variation mainly constrain the generalizability of observations from priming, and therefore call for a more systematic investigation of individual- and context-specific effects. Variation across time – which derives from the UBCL principle that every single episode of language use can have a permanent impact on linguistic knowledge – proves more problematic: It entails that, by testing it, priming experiments are changing the very representational system they wish to investigate. In usage-based understanding, such training effects are far from trivial: They affect the very essence of a language system that consistently reconstitutes itself through change and adaptation (e.g., Schmid Reference Schmid2015).
In sum, UBCL-based interpretations of priming effects prove to be very different from those made by B&P based on different theoretical (pre-)assumptions (e.g., that syntax and semantics are separate, sect. 2, para. 3. They thus also yield a very different picture of what can be the function of priming in language research. This clearly illustrates that priming experiments cannot “discriminat[e] among … accounts” of linguistic representations (B&P, sect. 1.2, para. 10) – or at least cannot do so if used as a single method and if conducted and interpreted with specific preconceptions about the nature of linguistic knowledge in mind.
I fully agree with Branigan & Pickering (B&P) that it is high time psychologists, psycholinguists, and linguists join forces in investigating linguistic knowledge. I also agree that priming experiments can contribute to arriving at a better, cognitively more realistic understanding of language. Priming evidence, however, is not immune to theoretical bias. I will illustrate this by approaching the priming effects reported by B&P from the perspective of usage-based cognitive linguistics (UBCL) (e.g., Croft Reference Croft2001; Goldberg Reference Goldberg1995, Reference Goldberg2003, Reference Goldberg2006; Langacker Reference Langacker, Barlow and Kemmer2000, Reference Langacker2008; The Five Graces Group Reference Beckner, Blythe, Bybee, Christiansen, Croft, Ellis, Holland, Ke, Larsen-Freeman, Schoenemann, Ellis and Larsen-Freeman2009; Schmid Reference Schmid2015); that is, I will (re-)interpret them based on the following assumptions:
“Language is an integral part of human cognition” (Langacker Reference Langacker1987, p. 12);
Linguistic knowledge is highly dynamic and context-adaptive; it emerges from situated usage and is therefore potentially changed by every single usage experience;
Rules emerge from generalization processes; they take the form of schematic form–meaning pairs (constructions); that is, they are equivalent in nature and structure to specific lexical items or multi-word units. Consequently, grammar/syntax and lexicon form a continuum;
Knowledge of language is organized in a complex network of constructions. This network:
• Licenses redundancy, that is, contains both specific linguistic structures and their different degrees of abstractions, as well as both larger (e.g., sentence-level) and smaller constructions (e.g., single lexical items);
• Is characterized by multiple (types of) relations that the single constructions (as well as their form- or meaning-sides only) can hold to each other.
These UBCL principles suggest alternative interpretations of priming effects to those proposed by B&P.
Most centrally, they allow for structural priming to be interpreted differently than as evidence in support of the autonomy of syntax. In UBCL terms, the only relevant difference between structural and (more) lexical priming effects is that schematic rather than specific form–meaning pairs are primed. This interpretation allows for all instances of priming (including lexical priming) to be modeled in a unitary fashion: as based on the same mechanisms and as applying to different levels of the same network of linguistic knowledge. Because this network also contains semi-specific elements (e.g., [NP go PP]; see, e.g., Goldberg Reference Goldberg2003), the UBCL view of priming also readily accommodates lexical boost effects. Thus, from a UBCL perspective, the major potential of using priming to investigate linguistic knowledge does not so much lie in differentiating between syntactic and lexical processes and representations, but in identifying which level(s) of the constructional network language users actually and/or habitually access (see also Tomasello Reference Tomasello2006, p. 3).
This construction-based UBCL view also suggests that structural priming effects (just as lexical priming effects; see B&P, sect. 1.4, para. 3) might not occur only when “prime and target involve the same construction” (B&P, note 10), but also when the constructions involved in prime and target processing are (closely) related. The relations driving priming effects can be of different types (e.g., schema–instance, part–whole, similarity, contrast, etc.) and can variably hold between entire constructions or between their form- or meaning-sides only (e.g., Langacker Reference Langacker1987, p. 13). Findings of priming between idioms and formally parallel non-idiomatic utterances (Konopka & Bock Reference Konopka and Bock2009), therefore, do not contradict construction-based UBCL models of language (as claimed by B&P, note 10). The same accounts for priming between sentences in which different thematic roles and/or syntactic functions are formally realized in the same way (e.g., B&P, sect. 2.1, para. 2; Pickering & Ferreira Reference Pickering and Ferreira2008). Instead, from a UBCL perspective, such findings are potentially informative of the internal relational structure of linguistic knowledge and the way it is used in processing.
Priming experiments thus also prove a valuable tool for language research when approached from a UBCL perspective. However, some UBCL assumptions call into question what B&P define as one main benefit of structural priming – namely, that it “provides evidence that is directly informative about mental representation” (sect. 1.5, para. 1).
One of those assumptions is that information is represented redundantly in the language network. It suggests that very different kinds and combinations of constructions can be involved in the processing of what is – formally – the same utterance (e.g., Günther Reference Günther2016, p. 143; Stefanowitsch & Flach Reference Stefanowitsch, Flach and Schmid2016, pp. 105–107); that is, an utterance such as The bottle is on the table might variably be accessed as a single chunk or involve the integration of several more schematic and more specific constructions (e.g., [NP is on NP] + [the bottle] + [the table], etc.). It is thus hardly possible to draw reliable conclusions about which and how many constructions a speaker used when producing or comprehending a particular utterance in a priming experiment.
What complicates matters even further is that, on a usage-based view (e.g., Günther Reference Günther2016), the sets of constructions language users access during utterance production or comprehension are highly likely to differ among individuals and may vary even within individuals across external (situational, social, and linguistic) contextual conditions, as well as across time. Interindividual and cross-contextual variation mainly constrain the generalizability of observations from priming, and therefore call for a more systematic investigation of individual- and context-specific effects. Variation across time – which derives from the UBCL principle that every single episode of language use can have a permanent impact on linguistic knowledge – proves more problematic: It entails that, by testing it, priming experiments are changing the very representational system they wish to investigate. In usage-based understanding, such training effects are far from trivial: They affect the very essence of a language system that consistently reconstitutes itself through change and adaptation (e.g., Schmid Reference Schmid2015).
In sum, UBCL-based interpretations of priming effects prove to be very different from those made by B&P based on different theoretical (pre-)assumptions (e.g., that syntax and semantics are separate, sect. 2, para. 3. They thus also yield a very different picture of what can be the function of priming in language research. This clearly illustrates that priming experiments cannot “discriminat[e] among … accounts” of linguistic representations (B&P, sect. 1.2, para. 10) – or at least cannot do so if used as a single method and if conducted and interpreted with specific preconceptions about the nature of linguistic knowledge in mind.