A comparative study of parsing in English and Japanese

Date of Completion

January 1991


Language, Linguistics




This thesis explores the operating principles of the human sentence processing mechanism (HSPM) by focusing on the resolution of structural ambiguity in Japanese. A left-to-right on-line parser such as is usually assumed for English, would constantly face a considerable degree of ambiguity in Japanese due mainly to the fact that the language is head final. However, intuitively detectable garden paths are not as severe as this kind of model would predict. Two explanations are considered. (i) The parser delays analysis of the input string until it receives information relevant to the decision. This would explain the absence of garden paths. It is a safe but perhaps not efficient strategy because the relevant information is so delayed in Japanese sentences that storing much unstructured material could over-load memory. (ii) The parser misanalyzes the input frequently but easily corrects the initial analysis when the relevant information becomes available. This would solve the memory problem. But it presupposes some further explanation of why reanalysis should be easier for Japanese than for English.^ It is proposed here that the Japanese parser is of type (ii). It is capable of analyzing the input in two ways: left-to-right on-line fashion, and look-back fashion. It is argued that the operating mode selection is dependent on the availability of relevant information. In the absence of such information the parser makes a tentative decision based on general structural preference. When precise information becomes available, the parser revises previous decisions if necessary. It is suggested that the cost of revising a decision is a function of the parser's confidence in making that decision.^ It is argued to be unnecessary to postulate two distinct processors for Japanese. Both styles of decision making could come from the same mechanism, which is paced by the information input. In that case, the processor for English can also be seen as an instance of the same mechanism. Its superficial performance differs because the information availability is differently paced in English sentences. This is important because there is reason to expect that HSPM should be innate and hence universal. ^