Discussion Details
| Title: | Minimalism Challenge |
| Submitter: | Michael Hegarty |
| Description: | [Richard Sproat and Shalom Lappin (LINGUIST 16.1156)]: ''In fact, we
would be delighted if someone succeeds in meeting our challenge [of computationally modeling the development of the child's ability to parse sentences]. Such success would convince us that the P&P enterprise is, after all, a testable theory with genuine scientific content.'' [end quote] Most likely, Sproat and Lappin do not consider amenability to computational modeling to be a requirement for all scientific theories (for there are easy counterexamples to that), but for all syntactic theories which, like Principles & Parameters syntax, aim to account for how children acquire language. Specifically, they assume that providing a basis for a computational model of how to attain the ability to parse sentences is a threshold level of achievement for such a theory. But curiously, for all their concern with ''building a computational model that is able to do what every child is able to do,'' (16.1156) Sproat and Lappin say essentially nothing about what children actually do, that is, about the actually observed linguistic behavior of children through the process of language development. It's one thing to computationally model processes which achieve a final state of successfully parsing linguistic input based on what one takes to be the child's initial state, and quite another thing to provide a computational model of the development of child language which systematically accounts for the linguistic forms produced by children and the characteristic changes in those forms over time. Peter Hallman's post (16.1251) makes a point along these lines, arguing that, in order to meet the concerns of P&P theory, a model of child acquisition of syntax would not just attain a parsing capability equal to that of the child's final state, but it ''would behave like a language learner in its acquisition timeline and would fail to acquire languages that violate language universals.'' Work in computational modeling of language acquisition has certainly not completely ignored the problem of accounting for child behavior en route, as attested by connectionist claims to reproduce child errors. But the full range of developmental issues which occupy P&P theorists of child language acquisition do not seem to be dealt with by the work cited by Sproat and Lappin. A response on behalf of Sproat and Lappin would be that they are merely demanding something simpler and weaker, as a threshold condition for a theory of syntax to establish its scientific bona fides. Marc Hamann's post (16.1472) fits into this response: if P&P theorists are interested in the acquisitional process in its full glory, then they should be ready to meet Sproat and Lappin's challenge since, after all, it pertains to only a part of that whole. But this response assumes that scientific theories come neatly nested, with all theories in an area providing coverage of basic phenomena, beyond which theories become progressively more specific and specialized. That isn't always how it works. Qualitative physiology gives a high-level account of the functioning and possible dimensions of impairment of the body. It's observations have even been systematized in ways which provide a basis for computational modeling. Cell-level and molecular physiology deal with lower-level phenomena, providing precise and detailed accounts of particular low-level processes, which can be computationally modeled in their own right, but which do not always lend themselves to ready synthesis into a theory of the whole system. This is not meant to be an exact analogy to the acquisition of syntax, and I leave it to others to discuss which of P&P syntax and computational modeling of child language acquisition is more analogous to qualitative physiology, and which to cell and molecular biology. The analogy is merely meant to illustrate the fact that different scientific approaches to the ''same problem'' might be sufficiently different in scale and focus that neither of them can account for the easiest or most basic test cases of the other. As a result, researchers must make choices regarding their focus. I want to understand the growth of child language and the nature of language impairment in the same way in which P&P syntax and, increasingly, other frameworks help us understand cross-linguistic variation. In particular, I want to understand things like the progression through the optional infinitive phase mentioned by Seth Cable (16.1454), and the phenomena which I understand as reflecting growth in the child's capacity for representing functional categories, described below. P&P syntax provides tools for me to make progress on these problems; to my knowledge, work on computational modeling of the attainment of the ability to parse sentences does not. Things change, of course, and mathematically better defined theories of syntax such as HPSG are increasingly attractive and effective instruments for the sort of comparative investigation which has long been the forte of P&P syntax, and computational modeling may eventually become detailed enough to answer all my questions about the acquisitional process. In the meantime, I am no more impressed by the insistence that P&P do what Sproat and Lappin and Hamann want it to do than I would be by the insistence that cell and molecular biology demonstrate their bona fides by accounting for whole organ or whole organism function in line with results of qualitative physiology, or that qualitative physiology match the explanatory detail of cell and molecular biology at the microlevel. A case-study is helpful to break down some stereotypes concerning P&P syntax, including the supposition that all P&P work is straightforwardly committed to some identifiable form of the innateness hypothesis. Forgive the element of self promotion which follows, but I must choose an example which is ready to hand. In Hegarty (forthcoming), I report on a study of child acquisition of functional elements which found that there is a stage at which children have acquired individual classes of functional elements, including complementizers, manifestations of inflection, and negation, and display them one or two at a time within a clause, but, despite more than adequate phrase structure complexity (and working memory and processing capacity) they don't display three functional elements together within a clause. The pattern is general: there are successive stages at which the child displays a maximum of (successively) 0, 1, 2 and finally 3 functional categories within a clause. I interpret these results as a consequence of growth in the representational resources of the child's grammar. The pattern bears some similarity to that predicted by the bottom-up structure building theory of the acquisition of functional categories (Vainikka 1993/94, Guilfoyle and Noonan 1992), but is distinct from it. I discovered this pattern by virtue of working within a feature-based formulation of the syntax of functional categories, a part of P&P theory with its roots in Minimalist syntax. The framework served as an engine of discovery of this empirical generalization, and it provides theoretical machinery to express it neatly. But other frameworks and approaches will be able to accommodate the result. (Other hypotheses about functional category acquisition fail to predict the results I found, but the failed predictions do not accrue to other frameworks of syntactic theory.) I look forward to research within any framework or any approach to the development of clause structure supporting functional elements which sheds light on what I identify as the growth of representational resources of the child's grammar, what those resources are and how they should best be characterized, and what this means for the innateness hypothesis and strong continuity approaches to functional category development (which I see my results as challenging, at least in some measure). I also look forward to developing proposals regarding the neurological instantiation of this growth in representational capacity along the lines of proposals by Wakefield and Wilcox (1995) and Wilkins and Wakefield (1995) -- which addressed the neurological instantiation of Radford's (1990) theory of the maturation of functional categories -- but in light of more current neurolinguistic ideas of the sort surveyed in the issue of Cognition edited by Hickok and Poeppel (2004). The body of known generalizations about child language development has grown with the line of work of which mine is a part, our understanding of the process has grown, new directions for research have been uncovered, and our capacity to decide between competing fundamental hypotheses about acquisition, innateness versus maturation, for example, or to refine those hypotheses in response to facts about child language acquisition, has grown. I don't see any reason why this work should fail to count as science because it hasn't yielded improvements in computational modeling of the child's ability to parse sentences. Now we can return to the heart of Sproat and Lappin's argument for the legitimacy of their challenge: [Sproat and Lappin (16.1439)] ''Regarding B, we remind readers that humans do assign structure to sentences, that assigning structure to sentences is surely a part of what syntax is about, that humans acquire this knowledge as part of language acquisition, and that P&P claims to provide an explanation of how this is achieved. So we are at a loss to understand why inducing a large-scale working parser from sample data is not a valid test of P&P.'' [end quote] I hope to have provided enough background to make clear the central problem with this apparently simple and straightforward line of reasoning. No one person speaks for P&P theory. If there is someone who claims to have an explanation of how humans acquire the knowledge to assign structure to sentences, then Sproat and Lappin's challenge might be rightfully directed at that person. But I and the researchers whose work I have cited make no such claim. We claim to have accounts of some central aspects of the development of the functional category system. More generally, we have accounts of aspects of child syntax, and aspects of the development of child syntax which, as far as I know, other theories have not even noticed. But we don't claim to have an explanation of how humans acquire the ability to parse sentences. For my own part, I don't even know if this development is accurately characterizable as an increase in knowledge, as Sproat and Lappin understand it, which seems to assume that the child's relevant cognitive structures and capacities are fully in place, simply waiting to iterate a learning procedure to convergence. For this reason, it's not even clear to me that existing computational models of how humans acquire syntax actually model the development of knowledge of language. One of the things I hope to get out of work on child language acquisition is a better understanding of what syntactic knowledge is, along with clarification regarding its genesis (innateness versus alternatives). ''What would count as a valid test of such a theory?'' is a common refrain in this discussion. Well, making predictions based on such theories, checking the data, and seeing which theory (or theories) best accords with the data sounds like a valid test. This is what I did with existing theories of functional category acquisition with regard to the development of the functional category system, given the child's inventory of expressed functional elements and level of phrase structure complexity. Why isn't this a valid test of a theory in progress? The amenability to computational modeling of the sort Sproat and Lappin demand is by no means a worthless quality for a theory to have, as Emily Bender's contribution to the discussion (16.1454) makes particularly clear. If everything else is equal, it is worth striving for. But it doesn't seem reasonable to regard it as the sole or primary measure of the scientific value of a syntactic theory. Michael Hegarty Department of English and Program in Linguistics Louisiana State University References: Guilfoyle, Eithne, and Maire Noonan. 1992. Functional categories and language acquisition. Canadian Journal of Linguistics 37(2): 241-272. Hegarty, Michael. Forthcoming. A Feature-Based Syntax of Functional Categories: The Structure, Acquisition and Specific Impairment of Functional Systems. Mouton de Gruyter. Hickok, G. and D. Poeppel, eds. 2004. Cognition 92. Radford, Andrew. 1990. Syntactic Theory and the Acquisition of English Syntax: the Nature of Early Child Grammars of English. Oxford: Oxford University Press. Vainikka, Anne. 1993/94. Case in the development of English syntax. Language Acquisition 3(3): 257-325. Wakefield, Jennie, and M. Jeanne Wilcox. 1995. Brain maturation and language acquisition: A theoretical model and preliminary investigation. In Dawn MacLaughlin and Susan McEwen, eds., Proceedings of the 19th Annual Boston University Conference on Language Development, vol. 2, 643-654. Cascadilla Press. Wilkins, Wendy K., and Jennie Wakefield. 1995. Brain evolution and neurolinguistic preconditions. Behavioral and Brain Sciences 18: 161- 182. |
| Date Posted: | 11-May-2005 |
| Linguistic Field(s): |
Computational Linguistics
Linguistic Theories Language Acquisition Discipline of Linguistics |
| LL Issue: | 16.1506 |
| Posted: | 11-May-2005 |
