14129.gif

Structurally Constrained Parsing

structurally Constrained parsing

The Theory

A basic assumption within the nativist framework (Bever, Fodor, & Weksel, 1965; Chomsky, 1959, 1968/2012) is that the only plausible type of “learning” is the one of hypothesis testing (as opposed constructing general principles inductively): the confirmation or dis-confirmation of predetermined constraints. Jerome Bruner’s early experiments provide a paradigm case in this regard (Bruner, 1973; Bruner, Goodnow, & Austin, 1956). In their experiments the subject uncovers and sorts cards into two stacks, exemplars and non-exemplars of some propertyX that the subject has to “learn” on the basis of the evidence presented (e.g. satisfies X, does not satisfy X, etc.). Their data confirmed that there is no learning without a priori constraints on the kind of hypotheses that are actually tested and on their order of precedence. Similar arguments, noticeably, have also been made by staunch empiricists as well (Goodman, 1983), thereby confirming Hume’s observation ‘that the traditional inquiry into whether the judgment of projectability derives from habitual projections or the habitual projections from judgment of projectability […] rely much on the original hand of nature on which they [beasts] improve little or nothing by the longest of experience.” Thus it is not without merit to take this notion of hypotheses testing/confirmation in investigations into the process of language triggering2.

In asking what evidence exists for innate constraints on Language two phenomena (among others) readily come to mind. The first concerns children’s grammatical conservatism (comission, not omission, cf. Snyder, n.d.), while the second concerns the evidence of active search for constituents in specific predetermined locations in online sentence processing. In this section I will take a brief look at both phenomena as providing an empirical background for psychological nativism. In analyzing children’s use of language, Snyder (n.d.) notes that while such usage seems full of errors, careful quantitative analyses shows that children either make correct use of a rule or omit it altogether. But what children rarely do is make a mistake in applying a rule or apply an incorrect rule. Why so? Before we consider a plausible answer, let us consider another related and curious phenomena. Senghas and Kegl (1994a) famously observed that deaf and dumb children in Nicaragua, put together in collective communal schools by the Sandinista Revolutionaries, had automatically triggered a complex new sign language sans any input at all. Not only did this new language manifest all the complexities of natural languages, but in fact the researchers confirmed that NSL converges with other sign languages on prior known universals (see also Berent, 2013).Further, researchers have provided empirical evidence for the fact that children are, in fact, capable of triggering combinatoric systems without external input altogether (Goldin-Meadow & Yang, 2017). Said combinatoric systems, too, converge with what is already known about language universals. In light of such facts I submit that children, equipped with a restrictive definition of Language in terms of predetermined constraints on plausibility and possibility, merely trigger one from among a possible set of combinatoric systems. Charles Reiss (Reiss, 2003) proposes the following cognitive typology.

Charles Reiss (2014): A Cognitivist's Typology

Charles Reiss (2014): A Cognitivist's Typology

Given the arguments put forth by Yang and colleagues, and in light of the empirical evidence from NSL, I argue that the role of exposure is limited to allowing the child to take a shortcut of sorts – instead of randomly triggering one possible system to externalize, the child merely opts to externalize the system already in use in its environment. NSL and such cases illustrate, however, that in cases of emergency (such as the criminal deprivation seen in Genie the Feral Child, and unfortunate incidents such as those in Nicaragua) children are perfectly capable of triggering languages of their own.1 If this is so, then children’s grammatical conservatism makes perfect sense since the child, being a natural scientist (as Carl Sagan often observed), skeptically subjects the input it receives to the process of hypotheses confirmation, making use of only what fits its a priori expectations and suspending judgment on other forms. Comission, not omission!

As mentioned before, yet another set of observable evidence for a priori expectations for stimuli to conform to deterministic principles comes from studies of online processing of sentences. ERP research concerning whether such dependencies impact upon active sentence comprehension have shown, for instance, that the parser actively seeks satisfaction of antecedent-pronoun dependencies and that such searching is sensitive morphology only in syntactically licit positions (Pablos et al. 2015). Further the authors also illustrate that similar results obtain in both backward and forward dependencies (Pablos et al. 2018). Generative theories treat backward dependencies similarly as forward dependencies as structural hierarchy is the only important factor. Such ERP studies examine the strategies employed by the parser in online interpretation despite the different nature of the relation between the dependent element and its licensor – even though the exact nature of the dependencies is different, both dependencies are restricted by syntactic structure. In other words, in both types of dependencies, there are positions in which the licensor can occur and positions from which it is impossible to enter into a licensing relation with the licensee.

The Experiments

We propose a total of three experiments to tap into the extent to which a priori constrained syntactic hierarchies are actively employed in online sentence processing by adults, and to explore whether seperate levels of linguistic knowledge can be tested to be psychologically real. Ding et al. (2016) have already established that neural oscillations track hierarchies in distinct windows that correspond to gradually building up of linguistic hierarchies. We explore further to tease apart the effects that stem purely from syntactic hierarchies and those that are derivable from surface semantics. Finally, in an infant directed experiment I propose to investigate possible continuities between the infant and adult syntax networks in the brain.

In our first experiment, conducted in two sets, we investigate the binding phenomena with regard to cataphoric pronouns in set 1, i.e. where the pronoun linearly precedes the antecedent. The aim of this study is threefold: (i) to examine whether GMME (gender mismatch effect) is observed when the parser encounters the first potential/syntactically licensed antecedent of the cataphoric pronoun that does not match in gender. This would be an indication that the parser starts actively searching for a matching antecedent after encountering the cataphoric pronoun, even though the antecedent of the pronoun could, theoretically, be found outside of the sentence (i.e. the parser is seeking confirmation for a priori expectations), (ii) to examine if the search mechanism is modulated by grammatical constraints such as Principle C of the Binding Theory. For cases where co-reference may lead to Principle C violations, we predict no difference between the match and the mismatch conditions. We predict that an ERP component is elicited only for referential expressions that can legitimately establish a co-reference relation with the cataphoric pronoun,. First, No-Constraint conditions, which contain a possessive pronoun, in masculine or feminine form that matches or mismatches, respectively, in gender with the linearly first antecedent Lodewijk (masculine). Second, Principle C conditions, which contain a cataphoric nominative pronoun in masculine or feminine form that cannot co-refer with the referential expression in the embedded clause due to Principle C. In all conditions, the test sentences always contain a licit antecedent for the pronoun.

Relevantly, even if pronouns could have co-reference with an antecedent outside of the sentence, the availability of an antecedent in the same sentence guarantees that the pronoun-antecedent relation is resolved within the sentence. Feminine and masculine pronouns and referential expressions will be counter-balanced. Previous reading time studies found effects at positions immediately following the antecedent ( Yoshida et al., 2014).

Accordingly, we elect to include proper names (such as Vishnu Dandem) in our data to ensure the availability of a region immediately following the proper name that remains connected to the antecedent position. The theoretical contribution here is a challenging one, since it comes from phenomena that are conceptually the same (binding) but different in the realization (directionality) — if we find that the parser responds similarly to hierarchical relations, despite differences in the configuration of each tested structure, then it must be the case that the parser can extract general grammatical properties out of all specific inputs and that it can deduce the structural hierarchy behind the linearly presented input.

The final and third motivation behind this experiment is tested in our set 2 experiments. Here we investigate the effects of case marking in instances of backward control structures. In particular, we test whether Krivochen’s (Krivochen, n.d.) hypotheses that an extra positive cognitive effect can be found from the presence of distinctly materialized case-marked tokens in utterances judged to be well-formed relative to ‘degraded’ utterances lacking such tokens. In this view, LEXICON is composed of ∆ type-variables, whose Spell-Out depends exclusively on the syntactic context. The variable enters the derivation as a token-variable bearing a potential plausibility as far as its phonetic form and interpretation is concerned. When CNP subjects materialize as Inherent Case-marked arguments, the Backward Control structures are considered more acceptable.  Since the position where such inherent case is fixed by syntactic hierarchies, this provides for another oppurtunity to probe whether the parser only looks for evidence in predetermined positions.

 

1.png

In our third experiment we test the subjects’ linguistic knowledge at each level by selectively disrupting one level while maintaining the others intact. More specifically, we ask the subjects to detect either phonological, morphosyntactic or syntactic anomalies in pseudo-word sentences which contained only one type of anomaly for each level . This in principle allows one to focus selectively on syntactic processing rather than on the different amount of syntactic complexity, as done in cited works. A major problem is also overcome which is implicit in this type of experiment. In fact, when the syntactic level is disrupted, a potential semantic anomaly is also produced; thus, for example, if one says “all the eaten have chickens snakes” the anomalous syntactic structure also disturbs the semantic interpretation which would be impossible to reconstruct.The study consists of three experimental and one baseline conditions. Subjects are asked to covertly read sentences presented visually and, for the three experimental conditions only, to make acceptability judgments at the corresponding sentence-structure levels. The sentences all consist of pseudo-words (‘pseudo-sentences’), so as to neutralize the access to semantic components: this ‘Quasi-Malayalam’, devoid of any open-class word, but maintaining inflections and function-words, is employed in order to isolate the correlates of morphosyntactic and syntactic processing. According to the experimental tasks, anomalies will be introduced either at the phonotactic, the morphosyntactic or the syntactic level. Syntactic anomalies present sentences with wrong linear order but proper agreements. Morphosyntactic anomalies present sentences with proper word order but agreement errors. Phonotactic anomalies present sentences containing Italian illegal consonant strings. The choice of Malayalam as the target language is based primarily on two reasons; (a) the rich morphosyntactic and phonological nature of Malayalam grammar (Mohanan, 1984) and (b) existence of previous neurolinguistics studies in the language that act as both a template for stimuli design, and also for comparison of results (di Sculio, 2009)

Finally, in our final set of experiments we aim to close the discussion of structures and a priori constraints in the mind with a continuity argument. The aforementioned remarkable and complex cognitive functions observed in humans do not suddenly emerge in adulthood but are shaped gradually during a critical period of development. After centuries of considering infants’ minds to be blanks slates, research in cognitive development has repeatedly shown considerable cognitive competencies in specific domains merely hours after birth despite their highly limited motor behavior (Gervain, Berent, & Werker, 2012; Mehler et al., 1988). This set of early capacities projects human infants on a learning pathway beyond the pathways available to other animals. This pathway observed even in neonates already bears some of the hallmarks of patterns of learning and cognition in human adults – for instance, human noenates are sensitive to identity relations in phonological structures (Gervain et al., 2012).  In some domains, such as language, human infants are even better learners than adults. In other domains, such as numerical cognition, infants lag far behind adults in the extended process of developing knowledge of mathematics. This suggests that the neonate mind is prewired in certain specific ways, as all biological organisms are in order for them to make sense of their environmental niche (cf. Tinbergen, 1951 ), that directs future learning and determines, among other things, why only certain C-R properties characterize human language while others do not. Indeed, the human infant brain is the only known machine able to instinctively produce an algebraic computational cognitive system capable of discrete infinity — infants are sensitive to their native language prosody at birth (Mehler et al., 1988), infants’  rapidly become sensitive to word categories, storing the most frequent function words of their native language by 6 months (Shi et al., 2006b),7-month-old infants are sensitive to the abstract pattern underlying syllable triplets (Marcus et al. 1999) .The ability to detect algebraic patterns associated with the use of labels for categories of objects, again, points to the possibility that human infants might have access to symbolic representations and recursive rules that constitute human adults’ common use of symbolic systems (e.g., speech, numbers, writing codes, algebraic formulae). In light of the NB framework proposed by Embick and Poeppel it is reasonable to assume that our ‘‘symbolic minds’’ must also depend on species-unique aspects of human neural architecture, in particular the expansion of associative cortical areas and the development of new long-distance fiber tracts, such as the arcuate fasciculus (Rilling et al., 2008). We aim to test such continuity between infant and adult sapien brains by using near-infrared spectroscopy (NIRS) to measure neonate’s neural sensitivity to syntactic computations of the kind regularly attested in adults, specifically L1 word-order (Malayalam).If, in fact, we are hardwired for specific linguistic computations, then at least with regard to those tasks we should expect some observable similarity between the infant and adult brains in both structural and functional terms. Given that figuring out the word order of their languages is fundamental to subsequent (Dehaene-Lambertz, Hertz-Pannier, & Dubois, 2006; Gómez et al., 2014; Mehler et al., 1988), and more complex, syntactic computations it is reasonable to assume that word-order sensitivity is likely to be among the prewired abilities of the neonate mind.

1Such cases are not unknown in the history of Evolutionary Biology. For detailed discussions see works by Ernst Mayr, E.O. Wilson and Richard Dawkins (Dawkins, 1999, 2006, 2015; Mayr, 1982; Wilson, 1999, 2000).

2See Piatelli-Palmarini (to appear in Gleitman & Almeida eds., A Festschrift for Jerry Fodor) for a discussion of acquisition vs. trigerring/trigers.

References:

Berent, I. (2013). The phonological mind. Cambridge: Cambridge university press.

Bever, T. G., Fodor, J. A., & Weksel, W. (1965). On the acquisition of syntax: A critique of “contextual generalization.” Psychological Review, 72(6), 467–482. https://doi.org/10.1037/h0022697

Bruner, J. S. (1973). Going Beyond the Information Given. New York: Norton.

Bruner, J. S., Goodnow, J. J., & Austin, G. A. (1956). A Study of Thinking. New York: Wiley.

Chomsky, N. (1959). On certain formal properties of grammars. Information and Control, 2(2), 137–167.

Chomsky, N. (2012). Cartesian Linguistics (3 edition). Cambridge University Press.

Dawkins, R. (1999). The extended phenotype: the long reach of the gene (Rev. ed). Oxford ; New York: Oxford University Press.

Dawkins, R. (2006). The selfish gene (30th anniversary ed). Oxford ; New York: Oxford University Press.

Dawkins, R. (2015). Brief Candle in the Dark: My Life in Science. Random House.

Ding, N., Melloni, L., Zhang, H., Tian, X., & Poeppel, D. (2016). Cortical tracking of hierarchical linguistic structures in connected speech. Nature Neuroscience, 19(1), 158–164.

Gervain, J., Berent, I., & Werker, J. F. (2012). Binding at birth: The newborn brain detects identity relations and sequential position in speech. Journal of Cognitive Neuroscience, 24(3), 564–574.

Goldin-Meadow, S., & Yang, C. (2017). Statistical evidence that a child can create a combinatorial linguistic system without external linguistic input: Implications for language evolution. Neuroscience & Biobehavioral Reviews, 81, 150–157. https://doi.org/10.1016/j.neubiorev.2016.12.016

Goodman, N. (1983). Fact, Fiction and Forecast. Cambridge, MA. and London: Harvard University Press.

Krivochen, D. G. (n.d.). Revisiting the Evaluator: Derivations (and Learning) in OT.

Mayr, E. (1982). The growth of biological thought: Diversity, evolution, and inheritance. Harvard University Press.

Mehler, J., Jusczyk, P., Lambertz, G., Halsted, N., Bertoncini, J., & Amiel-Tison, C. (1988). A precursor of language acquisition in young infants. Cognition, 29(2), 143–178.

Pablos, L., Doetjes, J., & Cheng, L. L.-S. (2018). Backward Dependencies and in-Situ wh-Questions as Test Cases on How to Approach Experimental Linguistics Research That Pursues Theoretical Linguistics Questions. Frontiers in Psychology, 8. https://doi.org/10.3389/fpsyg.2017.02237

Pablos, L., Doetjes, J., Ruijgrok, B., & Cheng, L. L.-S. (2015). Active search for antecedents in cataphoric pronoun resolution. Frontiers in Psychology, 6. https://doi.org/10.3389/fpsyg.2015.01638

Reiss, C. (2003). Quantification in structural descriptions: Attested and unattested patterns. Linguistic Review, 20(2/4), 305–338.

Senghas, R., and J. Kegl. 1994a. Social Considerations in the Emergence of Idioma de Signos Nicaragüense (Nicaraguan Sign Language). Signpost. vol.7, no. 1, Spring, pp. 40–46.

Snyder, W. (n.d.). Children’s Grammatical Conservatism: Implications for Syntactic Theory, 20.

The Study of Instinct. N. Tinbergen. New York: Oxford Univ. Press, 1951. 228 pp. $7.00 | Science. (n.d.). Retrieved August 27, 2018, from http://science.sciencemag.org/content/115/2990/438.2

Wilson, E. O. (1999). Consilience: The unity of knowledge (Vol. 31). Vintage.

Wilson, E. O. (2000). Sociobiology. Harvard University Press.