We were encouraged by the broad enthusiasm for building machines that learn and think in more human-like ways. Many commentators saw our set of key ingredients as helpful, but there was disagreement regarding the origin and structure of those ingredients. Our response covers three main dimensions of this disagreement: nature versus nurture, coherent theories versus theory fragments, and symbolic versus sub-symbolic representations. These dimensions align with classic debates in artificial intelligence and cognitive science, although, rather than embracing these debates, we emphasize ways of moving beyond them. Several commentators saw our set of key ingredients as incomplete and offered a wide range of additions. We agree that these additional ingredients are important in the long run and discuss prospects for incorporating them. Finally, we consider some of the ethical questions raised regarding the research program as a whole.

Within the cognitive sciences, most researchers assume that it is the job of linguists to investigate how language is represented, and that they do so largely by building theories based on explicit judgments about patterns of acceptability – whereas it is the task of psychologists to determine how language is processed, and that in doing so, they do not typically question the linguists' representational assumptions. We challenge this division of labor by arguing that structural priming provides an implicit method of investigating linguistic representations that should end the current reliance on acceptability judgments. Moreover, structural priming has now reached sufficient methodological maturity to provide substantial evidence about such representations. We argue that evidence from speakers' tendency to repeat their own and others' structural choices supports a linguistic architecture involving a single shallow level of syntax connected to a semantic level containing information about quantification, thematic relations, and information structure, as well as to a phonological level. Many of the linguistic distinctions often used to support complex (or multilevel) syntactic structure are instead captured by semantics; however, the syntactic level includes some specification of “missing” elements that are not realized at the phonological level. We also show that structural priming provides evidence about the consistency of representations across languages and about language development. In sum, we propose that structural priming provides a new basis for understanding the nature of language.

Structural priming is a useful technique for testing the predictions of linguistic theories, but one cannot conclude anything definitively about the shape of those theories from any particular methodology.

Although structural priming is often the most suitable paradigm, it sometimes misses effects that are detected by more sensitive acceptability-judgment tasks, thus yielding incorrect conclusions. For example, Branigan & Pickering's (B&P's) claim that “syntactic representations do not contain semantic information” (sect. 2.1, para. 2), while supported by structural-priming studies of the passive, is undermined by an acceptability-judgment study of this construction.

With a particular reference to second language (L2), we discuss (1) how structural priming can be used to tap into L2 representations and their relationships with first and target language representations; and (2) how complex networks additionally can be used to reveal the global and local patterning of L2 linguistic features and L2 developmental trajectories.

Structural priming is a sufficient but not a necessary condition for proving the existence of representations. Absence of evidence is not evidence of absence. Cognitive science relies on the legitimacy of positing representations and processes without “proving” every component. Also, psycholinguistics relies on other methods, including acceptability judgments, to find the materials for priming experiments in the first place.

In our commentary on Branigan & Pickering (B&P), we start by arguing that the authors implicitly adopt several assumptions, the consequence of which is to make further claims necessary and/or sufficient. Crucially, the authors assume the existence of discrete units at various levels of linguistic granularity that then must be operated upon by combinatorial mechanisms and rules (i.e., decomposition/recomposition). They further argue that structural priming provides a powerful tool to study abstract, structural representations. We provide evidence that priming effects in production are characterized better as graded than as all-or-none and that priming need not arise from a mechanism that (re)activates a shared but abstract internal structure.

Branigan & Pickering (B&P) argue successfully that structural priming provides valuable information for developing psychologically plausible syntactic and semantic theories. I discuss how their approach can be used to help determine whether partially grammaticalized constructions that have undergone semantic change also have undergone syntactic reanalysis. I then consider cases in which evidence from priming cannot distinguish between competing syntactic analyses.

A critical flaw in Branigan & Pickering's (B&P's) advocacy of structural priming is the absence of a theory of priming. This undermines their claims about the value of priming as a methodology. In contrast, acceptability judgments enable clearer inferences about structure. It is important to engage thoroughly with the logic behind different structural diagnostics.

Assuming that linguistic representation has been studied only by linguists using grammaticality judgments, Branigan & Pickering (B&P) present structural priming as a novel alternative. We show that their assumptions are incorrect for cognitive-functional linguistics, exposing converging perspectives on form/meaning pairings between generativists and cognitive-functional linguists that we hope will spark the cross-disciplinary discussion necessary to produce a cognitively plausible model of linguistic representation.

Usage-based cognitive linguistic (UBCL) theories offer a unifying interpretation of the different (structural vs. [more] lexical) priming effects reported by Branigan & Pickering (B&P), and they provide an ideal basis for explaining contextual influences on priming. However, they also call into question B&P's claim that priming “provides evidence that is directly informative about mental representation” (sect. 1.5, para. 1).

Acceptability judgments are no longer acceptable as the holy grail for testing the nature of linguistic representations. Experimental and quantitative methods should be used to test theoretical claims in psycholinguistics. These methods should include not only behavior, but also the more recent possibilities to probe the neural codes for language-relevant representations.

The target article's call to end reliance on acceptability judgments is premature. First, it restricts syntactic inquiry to cases were a semantically equivalent alternative is available. Second, priming studies require groups of participants who are linguistically homogenous and whose grammar is known to the researcher. These requirements would eliminate two major research areas: syntactic competence in d/Deaf individuals, and language documentation. (We follow the convention of using deaf to describe hearing levels, Deaf to describe cultural identity, and d/Deaf to include both. Our own work has focused on Deaf signers, but the same concerns could apply to other deaf populations.)

Structural priming data are sometimes compatible with several theoretical views, as shown here for three key theoretical claims. One reason is that prime sentences affect multiple representational levels driving syntactic choice. Additionally, priming is affected by further cognitive functions (e.g., memory). We therefore see priming as a useful tool for the investigation of linguistic representation but not the only tool.

As Branigan & Pickering (B&P) argue, structural priming has important implications for the theory of language structure, but these implications go beyond those suggested. Priming implies a network structure, so the grammar must be a network and so must sentence structure. Instead of phrase structure, the most promising model for syntactic structure is enriched dependency structure, as in Word Grammar.

Despite enthusiastic agreement that experimental data are directly relevant for determining grammar architecture, we present one main objection to the conclusions that the authors draw from their results: The data are perfectly compatible – in fact, much more in line – with an alternative that does not rely on syntactic representations. Instead, it is processing actions whose activation for comprehension/production explains intra-/inter-speaker priming.

In their target article, Branigan & Pickering (B&P) briefly discuss bilingual language representation, focusing primarily on cross-language priming between single-language sentences. We follow up on this discussion by showing how structural priming drives real-life phenomena of bilingual language use beyond the priming of unilingual sentences and by arguing that B&P's account should be extended with a representation for language membership.

The nature of mental representations of linguistic expressions in relation to the time course from intention to articulation is a major issue. We discuss Branigan & Pickering's (B&P's) proposal to use structural priming to tap into this process. We show that their interpretation of their findings cannot be maintained. We reinterpret these results and suggest a revision of their conclusions.

Like Branigan & Pickering (B&P), we agree that processing evidence is important for linguistic theorization; however, without much evidence of priming of hierarchical argument structure independent of linear ordering, the nature of “structure” in structural priming remains unclear. Consequently, it is an empirical question whether structural priming and acceptability judgments tap into cognitive processes of a similar nature.

We agree with Branigan & Pickering (B&P) that structural priming experiments should supplant grammaticality judgments for testing linguistic representation. However, B&P overlook a vast (corpus-)linguistic literature that converges with – but extends – the experimental findings. B&P conclude that syntax is functionally independent of the lexicon. We argue that a broader approach to priming reveals cracks in the façade of syntactic autonomy.