AN EVALUATION OF READING ASSISTANCE TOOLS IN FOREIGN LANGUAGE READING: SYNTACTIC PARSING AND MACHINE TRANSLATION

Computer assisted language learning uses NLP (natural language processing) techniques. For instance, foreign language reading can be assisted by NLP techniques such as syntactic parsing and machine translation. Syntactic parsing shows phrase boundaries of a sentence, and machine translation shows word translation and sentence translation. Previous studies (Ohguro 1993, Fuji et al. 2001) reported RA (reading assistance) effect on COMP (comprehension) of foreign language reading. Following these studies, this paper examined RA effect on COMP and RS (reading speed) for 102 EFL (English as a foreign language) learners whose native language was Japanese.
We assess RA effect in English texts, Japanese translation texts of the English texts, and texts with RA. The English texts are taken from TOEIC (Test of English for International Communications) textbooks. Texts with RA are the English texts glossed either with phrase boundaries, word translation or sentence translation. COMP is defined with the percentage of questions correctly answered in a text. RS is defined as the number of words read in a minute. RS is measured with a reading process monitoring tool that can record the reading time of each sentence (Yoshimi et al. 2005).
Among the three types of texts, English texts are regarded as the least readable texts for Japanese EFL learners, Japanese texts are taken as the most readable texts, and texts with RA are supposed to hold some readability between English texts and Japanese texts. Since sentence translation represents the meaning of English sentences in Japanese, sentence translation should show the strongest RA effect among the three types of RA. RA effect of word translation should follow that of sentence translation. The weakest RA effect should be observed in texts with phrase boundaries.
Text-level RA effect was examined with COMP and RS. The experimental results showed an unexpected readability ranking. Even though Japanese texts were found to be the most readable texts based on both COMP and RS, sentence translation showed RA effect only on COMP.
Sentence-level RA effect was examined with RS. Examination of sentence-level RA effect showed that RA effect appeared in less readable English sentences, while RA effect was not observed for more readable English sentences.
We further examined whether RA effect depended on EFL learners’ proficiency, i.e., beginner-level, intermediate-level and advanced-level. It was found that RA effect depended on learners’ proficiency. RA effect was stronger for less proficient learners. Advanced-level learners showed RA effect only on phrase boundaries regarding RS, intermediate-level learners showed RA effect only on sentence translation regarding COMP, and beginner-level learners showed RA effect on phrase boundaries and word translation regarding COMP and on sentence translation regarding COMP and RS.
Experimental results in text reading lead us to conclusion that RA effect varies by learners’ proficiency and reading strategies. Experimental results in sentence reading suggested that NLP techniques should be used for texts with low readable texts such as authentic texts. RA will help beginner-level learners in IR (intensive reading) and ER (extensive reading), intermediate-level learners in IR and advanced-level learners in ER.