Commentary on "The neural basis of predicate-argument structure", by James Hurford

Abstract: 76 words
Main Text: 1207 words
References: 88 words
Total Text: 1371 words

Do sensorimotor processes have reflexes in sentence syntax as well as sentence semantics?

Alistair Knott
Department of Computer Science
University of Otago
P.O. Box 56
New Zealand
(+64 3) 479 8299


Predicate logic has proved a very useful tool for the expression of theories of natural language semantics. Hurford's suggestion that predicate-argument structures mirror certain properties of the human sensorimotor architecture can be seen as an explanation of why this is so. While I support this view, I think that the correspondences which Hurford draws between linguistic and sensorimotor structures do not only involve natural language semantics, but include some elements of natural language syntax as well.

The connections which Hurford draws between structures in predicate logic and the architecture of the human sensorimotor system are very compelling. They provide a strong argument for the idea that the meaning of a sentence consists (at least in some part) in a fairly direct evocation of sensorimotor processes. This idea is already quite widespread in cognitive linguistics---Hurford's main contribution is to ground it very solidly in contemporary models of neuroscience and of linguistic structure. Nonetheless, the more detail that can be recruited in formulating the hypothesis that natural language semantics is grounded in sensorimotor structures, the more convincing it will be. For me, the most exciting contribution of the paper is that it paves the way for new research methodologies in both linguistics and sensorimotor psychology, to generate and test more detailed versions of the hypothesis.

My own interest is in extending Hurford's hypothesis to cover certain aspects of natural language syntax as well as natural language semantics. Hurford (2002) has cautioned against this, arguing that many aspects of syntax will be recalcitrant to this kind of sensorimotor grounding. However, I think that his position in the current paper requires some revision of his earlier argument.

For one thing, Hurford seems to have revised his conception of mental representations since his 2002 paper. In his earlier article, mental representations are emphasised as having no temporal structure: 'all parts of the representation of a remembered event are simultaneously present to the mind'. In his current paper, however, Hurford is at pains to emphasise the sequential nature of mental representations. The representation underlying a predicate-argument structure is clearly characterised as a sequence of two mental processes: first a direction of attention, and then (with implicit reference to this) an action of classification. Hurford's references to Ballard et al. (1995; 1997) reinforce this view: for these authors, mental representations are very largely defined in relation to an ongoing sequence of rapid directions of attention. There is certainly a nontemporal aspect to mental representations too, on Ballard et al.'s view and in Hurford's current article, since details of the most recently-attended-to objects are held in a short-term store, to facilitate re-attention. But this is certainly not to say that mental events have no temporal structure.

If mental representations involve sequences of directions of attention, one possible role for syntax is in capturing this temporal structure. I will begin by considering an aspect of syntax which Hurford (2002) singles out as being unlikely to reflect prelinguistic mental representations: the existence of a VP constituent, which groups the verb of a sentence with its grammatical object, but excludes the subject. For Hurford (2002), 'it seems unlikely that the structure of prelinguistic thought included a VP-like constituent which bracketed a 2-place predicate with one of its arguments, but not the other'. But I think the dynamic conception of mental representations in Hurford's current article opens up precisely this possibility.

One of the syntactic motivations for keeping the subject separate from the other arguments of the verb is the possibility of 'expletive' subjects; for instance, the subject of an existential sentence like 'There is a cup on the table'. According to a reasonably generic version of government-and-binding theory, verbs create ('project') syntactic positions for their semantic arguments to fill. The VP is basically defined as the ensemble of these positions. However, subject ([Spec,IP]) position is not projected by the verb; there is a separate requirement that all sentences must have a subject position. Further independent grammatical principles allow or require one argument of the verb to 'move' to subject position in some circumstances. In the sentence just given, no such movement has taken place; 'there' is a phonological expression of an empty subject position.

I suggest that the empty subject in an existential sentence can be thought of as denoting a bare action of attention, and that the constituents in the VP can be seen as denoting psychological processes which are deictically referred to this initial action of attention. For instance, the VP-internal NP 'a cup' can be thought of as contributing a pure operation of the classification system, occurring in the context set up by attentional capture. In a sentence with a transitive VP (for instance 'grabbed the cup') the verb can be seen as contributing an operation of action classification referred to this same attentional context. Passive sentences tell us that actions can be identified without their agents being classified as objects, but this is certainly consistent with evidence from psychology and neuroscience of a modular faculty for 'biological motion recognition' that operates independently from recognition of the agent as an object (see e.g. Johansson, 1973; Grossman et al, 2000). Interestingly, biological motion detection does nonetheless require focal attention (Thornton et al., 2002). The picture which emerges is one in which processes denoted by constituents in the VP are all deictically referred to the initial action of attention associated with subject position. In other words, Hurford's deictic conception of perceptual processing does show some promise as the basis for a sensorimotor characterisation of the subject-VP distinction.

One other syntactic phenomenon is worth mentioning, to extend the sensorimotor interpretation of subject and VP just given. Hurford argues in both the papers I am discussing that the distinction between definite and indefinite NPs should be attributed to the communicative role of language, rather than to its role in representing mental processes. But I think his account of sensorimotor structure allows a lot to be said about this distinction too. Hurford talks extensively about the notion of an object file: a representation in working memory of an object recently attended to, which serves to facilitate re-attention to the object during the course of performing a task. It is interesting to consider the different operations involved in creating and accessing these memory representations. Presumably, a new object file is created when an object is first perceptually established---i.e. after an episode of attentional capture. There are also operations when an existing object file is used 'top-down', to help reattend to an object. These two operations seem very suitable as the sensorimotor denotations of 'a' and 'the' respectively. 'The' is a presuppositional construction: to work out the semantic contribution of 'the cup', we must look for a salient referent with the property 'cup' in the discourse context, and the construction only has a denotation if we find one. 'A cup', on the other hand, introduces such a referent into the discourse context. Although I agree with Hurford that referring expressions in communicative contexts need to take into account the user's discourse model, maybe it could be fruitful to associate the set of referents in the discourse context with the set of object files. (The fact that only a small number of object files can be attended to simultaneously does not rule out the possibility that a larger set can be stored in short-term memory and used to guide some less detailed form of top-down attention.) This sensorimotor construal of definite and indefinite NPs is certainly consistent with my earlier suggestion about the subject-VP distinction. If existential sentences denote episodes of attentional capture, associated with the creation of new object files, then we expect to find indefinite NPs in these contexts and not definite ones, and indeed a sentence like 'There is the cup on the table' is ill-formed.


Grossman E, Donnelly M, Price R, Pickens D, Morgan V, Neighnor G and Blake R (2000). Brain areas involved in perception of biological motion. Journal of Cognitive Neuroscience 12(5):711--720.
Hurford, J (2002). The roles of expression and representation in language evolution. In A Wray (Ed.), The transition to language. Oxford: Oxford University Press.
Johansson G (1973). Visual perception of biological motion and a model for its analysis. Perception and Psychophysics, 14:201--211.
Thornton I, Rensink R, and Shiffrar M (2002). Active vs. passive processing of biological motion. Perception 31:837--853.