Comment on Arbib

Word Counts

ABSTRACT 60,

MAIN TEXT 1009,

REFERENCES 192

ENTIRE TEXT (TOTAL) 1327

Towards a Functional Neurophysiology of Argument Structure Mapping

Peter Ford Dominey

CNRS Institut des Sciences Cognitives

67 bd Pinel

69675 Bron, France

Tel: 33 (0)437911266

dominey@isc.cnrs.fr

isc.cnrs.fr

60 word ABSTRACT

Human language includes the ability to perform the mapping or transformation between semantic structures and the corresponding grammatical structures. Arbib has traced out an interesting trajectory of how human language could have evolved, but he appears to have neglected consideration of the functional and neural descriptions of these mapping processes.

The current comment provides the overview of a possible solution.

MAIN TEXT (with paragraphs separated by full

blank lines, NOT tab indents)

In Section 7 Arbib outlines the transition of case frames such as “Grasp(Leo, raisin)” – from an action-object frame to a verb-argument structure, and he claims that this is the crucial step in the transition from protolanguage to human language. In this context, utterences will no longer correspond in a holistic manner to their meanings, but rather, the “noun like” and “verb like” components of the utterence will be extracted and mapped onto the corresponding components of the meaning representation. Thus, human language allows a capability for mapping an enormously rich diversity of sentences types to their corresponding meaning.

Arbib suggests that this ability – the crucial core that would include compositionality and recursive structure - developed through cultural evolution with no requirement for specialized neurophysiological structure. Where we would have hoped for more enlightenment on these issues, Arbib only breifly sketches the progression from a protolanguage to human language via cultural evolution and glosses over some of the most important questions. In holistic phrases, the mapping from utterence to meaning is monolithic. As suggested by Arbib in his analysis of Wray, as “words” become somehow liberated from these structures, slots are formed in the holophrase, which can be filled in by different arguements. This progressive liberation yields a functional distinction between open class elements that fill in the slots, and the underlying templates or grammatical structure that defines the configuration of slots. I previously pointed out (Dominey 2004) the analogy between this evolutionary development described by Wray (2000), and the parallel trajectory in child language development characterized in detail by Tomasello (2003), with a transition from holophrases via pivot structures and verb islands to abstract argument constructions.

The endpoint of the evolutionary and developmental trajectories is a state in which novel sentences can be parsed, and meaning extracted and mapped onto conceptual representations. This requires acquisition and exploitation of knowledge including lexical categories, grammatical morphology and phrase structure regularities in order to combine lexical semantics with potentially complex phrasal semantics. This represents a significant qualitative leap from protolanguage, or even from a simple fixed S-like V-like O-like system. Arbib simply tells us that “It might have taken many, many millennia for people to discover syntax and semantics in the sense of gaining immense expressive power by ‘going recursive’ with a relatively limited set of strategies for compounding and marking utterances.” He fails however to provide a functional and a neurophysiological characterization of how this could happen: How does abstract argument structure mapping work? Does it rely on specialized neural machinery? What is the underlying neurophysiology?

In order to begin to understand the functional neurophysiology of flexible mapping from sentences to meanings, we have studied brain function during the processing of sentences and particular sensory-motor sequences (reviewed in Dominey et al. 2003). The important point is that both the sentences and the sequences require analogous forms of mapping: sentences require systematic mapping of grammatical onto semantic structure; and our sequences require systematic mapping of sequential structure within the sequence itself. Extending the analogy, in both cases, the particular transformation is signaled by special “closed class” or “function”elements. In language, closed class elements include prepositions like “to” and “by” that can help indicate, for example, whether an active or passive transformation is to be used. In sequences such as ABCxBAC and ABCyABC where A, B and C represent “open class” elements, the “function” symbols “x” and “y” indicate two specific and distinct transformations.

It has been observed that the processing of closed class function words, in particular those that indicate an upcoming transformation, elicits brain activation that results in a left anterior negativity (LAN) ERP profile 400-600 ms following the target word (e.g. Neville et al. 1992, Brown et al. 1999). How can Arbib explain this via cultural evolution? Did the “discovery of syntax” result in dissociable neurophysiological systems for processing open vs. closed class words, or was there already a non-linguistic correlate in place. In order to address this issue, Hoen and Dominey (2000) recorded ERP responses to the “closed class” elements or “function symbols” “x” and “y” while subjects observed and judged sequences of the format ABCxBAC and ABCyABC. Processing of the function symbols resulted in ERP profiles remarkably similar to the LAN observed for function words. This suggests that non-linguistic sensorimotor sequences also exploit the categorical distinction between function vs content elements, that was first evloved for sensory-motor function and then exploited for phrasal-conceptual function (see Dominey et al. 2003 for additional arguments).

We have implemented a sequence processing model based on this idea that simulates human linguistic and sequence learning behavior (Dominey 2000, Dominey et al. 2003, Dominey & Hoen in press). In the model, the dissociation of semantic and syntactic function reflects the obseration of this dissociation in Broca’ area (Newman et al. 2003). The structural transformation or mapping corresponds to a non-language-specific dorsal stream mechanism that culminates in the pars opercularlis (Ba 44) of the ventral premomotor cortex, relying on recurrent cortical networks and corticostriatal processing (Dominey et al. 2003) consistent with the procedural component of Ullman’s (2004) sentence processing model. In contrast, the integration of lexico-semantic content into this structure processing machinery corresponds to a language-specific ventral stream mechanism that culminates in the pars triangularis (Ba 45) of the ventral premotor area, consistent with the declarative component of Ullman’s model (2004).

Interstingly, the origin of the sequence transformation model was in the study of a simple form of analogical mapping in the sequence learning context (Dominey et al. 1995). The resulting suggestion is that within the sensory-motor domain, a form of analogical structure mapping developped which already exploited an open-class closed-class distinction, that could then be exploited in the “cultural evolution” phase in order to provide a flexible system for argument structure mapping.

ALPHABETICAL REFERENCE LIST (APA STANDARD)

Brown, C. M., Hagoort, P., & ter Keurs, M. (1999). Electrophysiological signatures of visual lexical processing: Open- and closed-class words. Journal of Cognitive Neuroscience, 11(3), 261–281.

Dominey PF (2000) Conceptual Grounding in Simulation Studies of Language Acquisition, Evolution of Communication, 4(1), 57-85.

Dominey PF (2002) From Holophrases to Abstract Grammatical Constructions in Development and Evolution, Evol Lang 2004, Leipzig.

Dominey PF, Hoen M, Lelekov T, Blanc JM (2003) Neurological basis of language and sequential cognition: Evidence from simulation, aphasia and ERP studies, Brain and Language, 86, 207-225

Dominey PF, Hoen M (in press) Structure Mapping (Ba 44) and Semantic Integration (Ba 45) in a Construction-Based Neurolinguistic Model of Sentence Processing, Cortex

Hoen, M., & Dominey, P. F. (2000). ERP analysis of cognitive sequencing: A left anterior negativity related to structural transformation processing. NeuroReport, 11(14), 3187–3191.

Neville, H. J., Mills, D. L., & Lawson, D. S. (1992). Fractionating language: Different neural subsystems with different sensitive periods. Cerebral Cortex, 2, 244–258.

Newman SD, Just MA, Keller TA, Roth J, Carpenter PA( 2003) Differential effects of syntactic and semantic processing on the subregions of Broca’s area, Cog. Brain Research, 16, 297-307

Tomasello, M. (2003) Constructing a language: Harvard University Press, Cambridge.

Wray A (2000) A protolanguage with no declaratives and no names. Evol Lang 2000, Paris.