Branigan & Pickering (B&P) present a case that linguistic theory should pay heed to the results of structural priming studies. I can agree with this wholeheartedly. Since the pioneering work of Bock (Reference Bock1986), structural priming has provided interesting evidence for the construction of linguistic representations as part of the process of sentence generation and understanding. B&P are somewhat ambivalent on the question of whether linguistic theory should pay heed only to structural priming (as seems implied in the abstract) or simply add it to the repertoire of methodologies for testing the predictions of theories. If simply an addition to the repertoire, again, I am in thorough agreement.
However, B&P draw unwarranted conclusions about the content of linguistic theory on the basis of their discussion of structural priming. B&P mention Chomsky's (Reference Chomsky1981) point that one should look carefully at the experiment, as well as the theory, when faced with a negative result in an experimental test of a theory. They write that this has been taken as an excuse for theorists to ignore experiments. It is, rather, an injunction to think about what the implications of the result mean. Often, theory is built on extremely solid empirical premises, and if the experiment appears to disconfirm the theory, then one has to ask whether the problem lies with the premises, the theory, or the experimental design. In many cases, the experiment simply does not have enough sensitivity to overturn an otherwise solid empirical base and concomitant theoretical conclusions.
Let's take a hoary example:
-
1.
We saw the boy with the telescope. As any Lin101 student will tell you, this is structurally ambiguous. There are numerous avenues for theoretically modelling this fact about (1). One approach would be to have some rule system that licenses two distinct representations: in one, a PP attaches to NP; in another, it attaches to VP. The hypothesized rule system makes predictions about the behaviour of other sentences. If a PP can attach to VP or NP, then this rule system predicts (2), (3), and (4) independently. The grammaticality of these sentences (and countless others, of course) is further evidence that what was hypothesized to explain the ambiguity is correct.
-
2.
The boy with the telescope arrived.
-
3.
We arrived with the telescope.
-
4.
We saw the boy with the telescope with the telescope.
Combined with the unassailability of the basic facts, the theory's deductive complexity is what gives it its epistemic strength.
Now, let us look at what structural priming has to tell us about (1) and examples like it. In previous work, Branigan et al. (Reference Branigan, Pickering and McLean2005) showed that, when the PP attachment height is disambiguated by pictures presented to experimental subjects, those subjects will maintain the same PP attachment height for similarly structurally ambiguous new sentences – a case of structural priming in comprehension. There is a twist, however: Such priming takes place only when the verb is repeated. There is no significant effect when the newly presented sentence is structurally ambiguous in the same way but contains a different verb. So, what can we conclude from this? Because structural priming has no effect, the sentences presented with different verbs are not structurally ambiguous? Imagine no priming effect had been found at all. Would this mean the theory was wrong? No. It would mean only that structural priming was insufficiently sensitive. There is no legitimate inference from the failure of structural priming to claims about the content of the grammar.
For similar reasons, B&P's arguments that mainstream generative linguistic theory is incompatible with evidence from structural priming do not go through: The problem is that structural priming is not sensitive enough to pick up on independently verifiable syntactic distinctions. B&P motivate their claim by using, inter alia, the fact that unergative sentences (such as Lilly danced) prime unaccusative sentences (such as Lilly froze). Generative syntax takes these verbs to be in two distinct syntactic classes, with their subjects having distinct structural properties. The priming result looks incompatible with this claim, because it places them in the same class. There is linguistic evidence for this distinction internal to English, however: Unaccusatives lead to resultative readings for secondary predicates, while unergatives do not (compare Lilly froze solid with *Lilly danced tired, which has, at best, a depictive reading). There is also a great deal of cross linguistic evidence for the distinction, motivated by numerous syntactic differences, including the kind of auxiliary that the verb appears with, the possible positions for the subject noun phrase, and the behaviour of pronominal elements in clause structure (see Alexiadou et al. [Reference Alexiadou, Anagnostopoulou and Everaert2004]). There is also solid processing evidence for the distinction from different experimental paradigms: Friedmann et al. (Reference Friedmann, Taranto and Swinney2008), in an online cross-modal lexical priming experiment on native English speakers, showed priming effects attributable to the two hypothesized structural positions for subjects; Lee and Thompson (Reference Lee and Thompson2011) showed that fixation differences in eyetracking are sensitive to the unergative/unaccusative distinction. Does the absence of a structural priming effect mean we should ignore linguistic, cross-linguistic, and psycholinguistic evidence? Does it mean that we therefore should adopt a theory in which no unergative/unaccusative distinction is made? No, because there is a simple explanation for the disparity in results. Structural priming is not sensitive enough to capture this syntactic distinction. B&P's argument from structural priming against mainstream generative syntactic theory is too logically weak to carry the burden placed upon it.
B&P take the limitations of structural priming to tell us something about the shape of the grammar and the representations that it legitimates. There is no reason to think this, however. There is no argument that takes us from the particularities of any method to conclusions about the object being studied, and to think that there is is to confuse the nature of the world with our ways of trying to understand it.
You have
Access
Branigan & Pickering (B&P) present a case that linguistic theory should pay heed to the results of structural priming studies. I can agree with this wholeheartedly. Since the pioneering work of Bock (Reference Bock1986), structural priming has provided interesting evidence for the construction of linguistic representations as part of the process of sentence generation and understanding. B&P are somewhat ambivalent on the question of whether linguistic theory should pay heed only to structural priming (as seems implied in the abstract) or simply add it to the repertoire of methodologies for testing the predictions of theories. If simply an addition to the repertoire, again, I am in thorough agreement.
However, B&P draw unwarranted conclusions about the content of linguistic theory on the basis of their discussion of structural priming. B&P mention Chomsky's (Reference Chomsky1981) point that one should look carefully at the experiment, as well as the theory, when faced with a negative result in an experimental test of a theory. They write that this has been taken as an excuse for theorists to ignore experiments. It is, rather, an injunction to think about what the implications of the result mean. Often, theory is built on extremely solid empirical premises, and if the experiment appears to disconfirm the theory, then one has to ask whether the problem lies with the premises, the theory, or the experimental design. In many cases, the experiment simply does not have enough sensitivity to overturn an otherwise solid empirical base and concomitant theoretical conclusions.
Let's take a hoary example:
1. We saw the boy with the telescope. As any Lin101 student will tell you, this is structurally ambiguous. There are numerous avenues for theoretically modelling this fact about (1). One approach would be to have some rule system that licenses two distinct representations: in one, a PP attaches to NP; in another, it attaches to VP. The hypothesized rule system makes predictions about the behaviour of other sentences. If a PP can attach to VP or NP, then this rule system predicts (2), (3), and (4) independently. The grammaticality of these sentences (and countless others, of course) is further evidence that what was hypothesized to explain the ambiguity is correct.
2. The boy with the telescope arrived.
3. We arrived with the telescope.
4. We saw the boy with the telescope with the telescope.
Combined with the unassailability of the basic facts, the theory's deductive complexity is what gives it its epistemic strength.
Now, let us look at what structural priming has to tell us about (1) and examples like it. In previous work, Branigan et al. (Reference Branigan, Pickering and McLean2005) showed that, when the PP attachment height is disambiguated by pictures presented to experimental subjects, those subjects will maintain the same PP attachment height for similarly structurally ambiguous new sentences – a case of structural priming in comprehension. There is a twist, however: Such priming takes place only when the verb is repeated. There is no significant effect when the newly presented sentence is structurally ambiguous in the same way but contains a different verb. So, what can we conclude from this? Because structural priming has no effect, the sentences presented with different verbs are not structurally ambiguous? Imagine no priming effect had been found at all. Would this mean the theory was wrong? No. It would mean only that structural priming was insufficiently sensitive. There is no legitimate inference from the failure of structural priming to claims about the content of the grammar.
For similar reasons, B&P's arguments that mainstream generative linguistic theory is incompatible with evidence from structural priming do not go through: The problem is that structural priming is not sensitive enough to pick up on independently verifiable syntactic distinctions. B&P motivate their claim by using, inter alia, the fact that unergative sentences (such as Lilly danced) prime unaccusative sentences (such as Lilly froze). Generative syntax takes these verbs to be in two distinct syntactic classes, with their subjects having distinct structural properties. The priming result looks incompatible with this claim, because it places them in the same class. There is linguistic evidence for this distinction internal to English, however: Unaccusatives lead to resultative readings for secondary predicates, while unergatives do not (compare Lilly froze solid with *Lilly danced tired, which has, at best, a depictive reading). There is also a great deal of cross linguistic evidence for the distinction, motivated by numerous syntactic differences, including the kind of auxiliary that the verb appears with, the possible positions for the subject noun phrase, and the behaviour of pronominal elements in clause structure (see Alexiadou et al. [Reference Alexiadou, Anagnostopoulou and Everaert2004]). There is also solid processing evidence for the distinction from different experimental paradigms: Friedmann et al. (Reference Friedmann, Taranto and Swinney2008), in an online cross-modal lexical priming experiment on native English speakers, showed priming effects attributable to the two hypothesized structural positions for subjects; Lee and Thompson (Reference Lee and Thompson2011) showed that fixation differences in eyetracking are sensitive to the unergative/unaccusative distinction. Does the absence of a structural priming effect mean we should ignore linguistic, cross-linguistic, and psycholinguistic evidence? Does it mean that we therefore should adopt a theory in which no unergative/unaccusative distinction is made? No, because there is a simple explanation for the disparity in results. Structural priming is not sensitive enough to capture this syntactic distinction. B&P's argument from structural priming against mainstream generative syntactic theory is too logically weak to carry the burden placed upon it.
B&P take the limitations of structural priming to tell us something about the shape of the grammar and the representations that it legitimates. There is no reason to think this, however. There is no argument that takes us from the particularities of any method to conclusions about the object being studied, and to think that there is is to confuse the nature of the world with our ways of trying to understand it.