Information from the Lexicon Is Fundamental to Sentence Syntax - Essay Example

Grammatical information from the Lexicon is fundamental to sentence syntax As part of the process of comprehending speech or text, the language system lexicon must achieve two important goals. First, the recognition of the individual and the meaning that they convey and second, the linking of these words to form phrases which convey further combinatory meaning. Recent developments in psycholinguistic theorizing have led many to question this supposed dissimilarity between lexical and phrasal processing. Many have begun to ask: What if we also recognize the presence of phrases And what if the detection of these phrases triggers the conveyance of combinatory meaning Such assumptions might provide continuity within theories of language comprehension because the system would become a probabilistic pattern recognition device through and through, detecting linguistic events of various sorts at multiple levels. These assumptions might also change what is thought to be involved in lexical and phrasal processing. As we discuss below, these assumptions imply a notion of lexical processing that bears considerably more responsibility for the combinatory analysis of language. Psycholinguistics has not been alone in this focus on the lexical aspects of combinatory process. Syntactic theory has increasingly moved detailed combinatory information into the lexicon, where individual lexical items are associated directly with their syntactic combinatory options. The field of applied parsing in computational linguistics has also seen a shift toward lexicalization. Many have recognized the effectiveness of coding these syntactic options as tendencies. In doing so, statistical natural language processing systems have begun to be able to recover the grammatical structure of novel sentences with astonishing accuracy. These movements in linguistics and computational linguistics touch on many of the same issues that have given rise to the development of constraint based lexicalist theories of parsing in psycholinguistics. In ways similar to the statistical NLP system, these theories propose that the recognition of a word involves the relative activation of detailed grammatical options, which are used to guide further combinatory processes. As a result, the frequency based activation of lexical alternatives becomes the basis for the resolution of many syntactic ambiguities. Some of the best support for this view has come from on-line studies of reading, which have shown that the sub categorization and thematic role tendencies of individual verbs can guide the resolution of local syntactic ambiguity. For instance, Garnsey examined readers' abilities to resolve temporary syntactic ambiguities involving classic direct object/sentence complement ambiguity. The use of lexical priming techniques in psycholinguistics has arguably been a highly effective tool for studying both the content of lexical representations and the time course with which such information is activated. For example, priming has been used to map the time course of activation of phonological and orthographic information during word recognition as well as the activation of the alternative meanings of ambiguous words. In most cases, however, the tasks used in these studies to measure participants reaction to target words are unlikely to be influenced by any hypothesized co activation of sub categorization or thematic role information, because such information is typically not relevant to successful execution of these tasks. Moreover, tasks that are sensitive to this sort of information, such as collection of reading times on individual words during sentence comprehension, have not been amenable to lexical priming techniques because the introduction of consciously perceived prime word, mid sentence, would catastrophically disrupt the ongoing comprehension of the sentence as a whole. In the early 1990's, however, Rayner and colleagues introduced a covert lexical intervention technique, dubbed fast priming, which allowed for the study of lexical priming effects during the free reading continuous text. In these studies, the eye movements of readers were tracked and used to contingently update the presentation of text on the computer screen. When the eyes landed on a target word, a prime word was presented for approximately thirty ms, followed by the target word. Although such events were typically perceived as a flicker, it was found that reading times were systematically influenced by the orthographic, phonological, and even semantic relationships between the prime target pair. Taken together, these data are highly consistent with theories of word recognition that allow for the parallel activation of orthographic, phonological, and semantic information associated with a letter string. A clear prediction of lexicalist parsing is that word recognition also includes the activation of detailed combinatory information, in the form of possible complements that a word may take. It is this activation process that out to influence the relative availability of alternative syntactic analyses. If this is the case, the syntactic preferences of a briefly presented prime word ought to have direct impact on a reader's parsing preferences of a syntactically ambiguous phrase. Trueswell and Kim tested these predictions in a series of experiments using a self paced reading version of fast priming. In the study, they examined how the fast priming of verbs can influence the way in which readers process sentences containing the direct object/sentence complement ambiguity. Like Garnsey, Trueswell and Kim compared the reading times of temporarily ambiguous sentences to unambiguous versions that included a "that". Increased reading times at the disambiguating phrase were taken as signs of misanalysis of the ambiguous NP as the direct object of the verb. Prior to reading each sentence, the words of the sentence were masked with each character in a word covered by an equal sign. Each press of the button uncovered a word and replaced the previous word with equal signs. On critical trials however, when the participant reached the matrix verb a prime word was displayed in its place for thirty nine ms. The prime verb was then replaced by the target verb which remained in the screen until the next press of the button. This event was typically perceived as a flicker on the screen, with the participants reporting in a detailed post experiment questionnaire that they rarely identified any prime words. Tow different types of prime words were compared which had been selected on their argument taking properties as measured from a separate sentence completion study. Primes were verbs that strongly prefer direct object and do not permit a sentence complement. SC primes were verbs that strongly prefer a sentence complement and rarely use a direct object. If the recognition of a verb includes the activation of its possible argument structures we might expect that the argument preferences of the flicker would influence the size of the garden path effect; DO primes should induce a large garden path effect whereas Sc primes should reduce the garden path effect. Indeed the experiments showed the expected pattern of priming. In particular Truewell and Kim observed that the processing difficulty usually found in the disambiguating region of ambiguous sentences was significantly decreased when the matrix verb had been primed with a SC prime as compared to when it had been primed with a Do prime. That is difficulty that stems from readers erroneously committing to a direct object analysis, due to their detecting the implausibility of the noun as the direct object and having to revise this syntactic commitment, was far less likely to occur when the prime provided argument structure information that could help the reader avoid this misinterpretation. Crucially the priming had its influence only on the ambiguous conditions and not the unambiguous conditions, implicating the primes' influence on avoiding the garden path and not the general fit of the prime into the sentence. One other observation from Truewell and Kim is also worth mentioning, because we will be comparing it to the findings of the first experiment. In both of their experiments they observed a thirty ms effect of priming at the verb, SC primes being faster than DO primes. This effect may suggest a partial role for semantic priming in this process, a point we return to later in this paper. In sum the Truewell and Kim findings show that garden path effects can be considerably mitigated by a briefly displayed prime verb, even though the primes were rarely identifiable to the readers. Prime verbs that prefer to take a sentence complement reliably reduced the garden path typically associated with ambiguous sentence complement constructions. Moreover, the timing of these primes as well as their subjective perception by participants strongly suggests that verb combinatory information of this sort is automatically activated during word recognition. The Truewell and Kim results contribute to a debate on the lexicon's role in comprehension that has focused heavily on the combinatory properties of verbs. This focus on verbs makes sense given the wide range of research assuming an anchor like status of verbs in the syntactic and thematic organization of language. Although verbs may be obvious starting point for the study of lexically based combinatory knowledge, lexicalist proposals clearly hypothesize that other types of words share the burden of combinatory knowledge representation. Perhaps the simplest reason for an account that extends beyond verbs alone is that grammatical phenomena suggest it. For instance, nouns like opinion can, like the verb accepted combine with a finite clause. If the knowledge that drives combinatory processing is lexically generated then nouns like opinion ought to encode knowledge about the complements they can occur with. This same construction illustrates another reason to extend the inquiry beyond verbs. The syntactic similarity between nouns like opinion and verbs like accepted suggests an underlying connection between their lexical representations. Specifically the representations of words like opinion and accepted may contain shared components even though the words belong to different basic grammatical categories. That is knowledge of sentential complements is encoded by overlapping distributed lexical representations. Studies have found that fast priming effects generalize beyond verbs to another part of the lexicon, nouns. The short lived processing of Sc bias noun primes influenced the syntactic processing of the host sentences in a manner that supports the sentential complement interpretation. The claims that processing of Sc bias nouns activates predictive knowledge about sentential complements. The representations that encode this knowledge are shared with sentential complement verbs and their pre activation influences the recognition of the target verb. By supporting the sentential complement sense of the ambiguous target verb, the priming pushes the processing system toward the correct grammatical analysis of the sentence. The similarity of abstract and concrete prime conditions constrains the interpretation. The possibility is that priming effects might arise from semantic abstractness in the primes. However, abstract primes behaved differently from SC bias primes and indistinguishably from concrete primes. Thus specifically combinatory knowledge is implicated over simple abstractness. It is intriguing to note that many of the abstract primes were even capable of taking complements of some sorts. Thus it appears that something quite specific is at work here. These results demonstrate that nouns can project syntactic structure in a way that plays an active role in the guidance of sentence structure in a interpretation. This contribution of noun argument structure is expected under lexicalist proposals given the need to specify combinatory information based on words other than verbs. For example, if processing is to be incremental in verb final languages syntactic hypotheses must be based on evidence stemming from the distributional probabilities of preverbal nouns and their morphological markers. These results also demonstrate that the representations of nouns and verbs are in some cases related through the sharing of representational mechanisms. The sharing of this representational machinery is the basis of the observed priming effects. This may reflect a general and powerful principle of the organization of grammatical knowledge: It is distributed across representational mechanisms which are shared by many elements of the lexicon, often across category boundaries. The principle of distributed knowledge representation is a fundamental claim if constraint based proposals in language processing and has been proposed specifically for lexically based parsing. An experiment, proportion of instrument actions as a function brought about mixed results. The results of this experiment are in many ways similar to the earlier fast priming studies; the combinatory preferences of an unattended prime word influenced ongoing parsing decisions involving an attended target sentence. Clearly as one might expect similar lexically specific parsing operations are at work in auditory language comprehension as in written language comprehension. Properties of this experimental task however as well as properties of stimuli used in the study permit us to make observations that go beyond merely replicating the reading findings in a different modality. For instance, the act out task that is part of the visual world technique allows one to examine in some detail the interpretation that speakers assign to each target utterance. In this regard, it is interesting to note that priming effects appeared to be restricted to the argument preferences of the primes and not to other aspects of the prime verb meaning such as the verbs core meaning. In particular inspection of the video record revealed no cases of participants acting out the prime verb rather than the target verb. Moreover there appeared to be no blending of the prime and target along these lines. The prime target pair did not result in fastidious turning. There is no reason to expect that the effects of a prime word should be restricted to the grammatical analysis of the sentence unless such primes somehow tap only implicit language operations as the grammatical analysis of the sentence. At the moment such a conclusion would be more speculation but clearly future research with this technique should systematically analyze the contribution of various event features denoted by these verbs. It is also worth noting that properties of the primes used in this experiment may also speak to the relative contribution of verb specific syntactic and semantic preferences to parsing decisions. Verbs were selected on the basis of sentence completion norms, in which participants were to complete sentence fragments. Completions were categorized on semantic grounds rather than on the basis of syntactic properties, in that only instrument and non-instrument completions served as categories. For instance, a verb was highly regarded to take an instrument role if there were a high number of prepositional phrase completions containing instruments. Sentences that contained VP attachments or NP attachments were categorized as non-instruments. This categorization scheme resulted in a large semantic difference between how likely certain verbs took instruments or how likely they were not to take instruments and this is how the prime types were chosen. Contrastingly inspection of gross syntactic differences in continuations for these verbs based on whether they were likely to take VP attachments or not resulted in very little difference in syntactic preferences. Work Cited Allopenna, P. D., Magnuson, J. S., & Tanenhaus, M. K. (1998). Tracking the time course of spoken word recognition using eye movements: Evidence for continuous mapping models.Journal of Memory and Language, 38(4), 419-439. Argaman, V., & Pearlmutter, N. J. (2002). Lexical semantics as a basis for argument structure frequency biases. In P. Merlo & S. Stevenson (Eds.), Sentence Processing and the Lexicon:Formal, Computational, and Experimental Perspectives. Amsterdam: John Benjamins. Britt, M. A. (1994). The interaction of referential ambiguity and argument structure in the parsing of prepositional phrases. Journal of Memory and Language, 33(2), 251-283. Carlson, G. N., & Tanenhaus, M. K. (1988). Thematic roles and language comprehension. In W. Wilkins (Ed.), Syntax and Semantics, Vol. 21: Thematic Relations (pp. 263-289). London: Academic Press. Cherry, C. (1953). Some experiments on the recognition of speech with one and with two ears.Journal of Acoustical Society of America, 25, 975-979. Read More