Three English Learner Assistance Systems Using Automatic Paraphrasing Techniques

Three English Learner Assistance Systems Using Automatic Paraphrasing Techniques Masaki Murata and Hitoshi Isahara National Institute of Information and Communications Technology 3-5 Hikaridai, Seika-cho, Soraku-gun, Kyoto 619-0289, Japan {murata,isahara}@nict.go.jp Abstract We developed three systems based on automatic paraphrasing techniques to help En- glish learners and English-language beginners. One system extracts personal error patterns in the user’s English usage. The second transforms English sentences contain- ing the letters “l” and “r” into sentences containing fewer instances of these letters, which Japanese people have trouble pronouncing properly in English. This system could be used, for example, to transform a draft of a presentation that a Japanese speaker was to present to an audience. The third is an annotation system that pro- vides definition sentences of difficult English words, making them easier to understand. We believe that these systems will be useful both for learners of English and in studies on second-language acquisition. 1 Introduction Studies on paraphrasing (IWPT, 2001; Murata and Isahara, 2001) are relevant to a range of research topics including sentence generation, summarization, and question-answering in natural language processing (Katoh and Uratani, 1999; Takahashi et al., 2003). Several techniques have been constructed for paraphrasing. We used automatic paraphrasing techniques based on natural language processing to develop three learner-assistance systems for English language learners and beginners. One system extracts an individual’s personal error patterns in using English. The second transforms English sentences containing the letters “l” and “r” into sentences with fewer instances of these letters, which Japanese people have trouble pronouncing properly in English. The third system transforms difficult English words into easier ones by providing definitions of the difficult words that clarify their meaning. These systems and studies using them may also have applications in studies on second language acquisition (Dulay et al., 1982; Larsen-Freeman and Long, 1991; Granger, 1998). 2 Extraction of personal error patterns in using English We investigated methods of automatically extracting differences from pairs of texts that cor- respond to each other within the domain of paraphrasing studies. One study examined the extraction of differences between spoken and written language and transformation rules for au- tomatic translation from written to spoken language by matching presentations at academic conferences and the papers corresponding to them (Murata and Isahara, 2002a). Another study focused on extracting rewriting rules and synonyms in patents by matching the patent claim and its embodiment, both of which have the same meaning (Murata and Isahara, 2002c). In this study, we extracted an individual’s personal English error patterns by matching a paper before proofreading with the corresponding paper after proofreading. (The proofreading was performed by native speakers.) We used 80 papers by a first author to extract the author’s PACLIC 18, December 8th-10th, 2004, Waseda University, Tokyo - 299 - Table 1: Examples of personal English errors Before proofreading After proofreading Frequency “the” ⇒ φ 393 φ ⇒ “the” 357 “a” ⇒ “the” 146 “a” ⇒ φ 122 φ ⇒ “a” 120 “the” ⇒ “a” 116 “is” ⇒ “was” 68 φ ⇒ “of” 56 …

Content truncated to 500 words