A new diagnostic system for J-E translation ILTS by global matching algorithm and POST parser

MT Summit VII Sept. 1999 A New Diagnostic System for J-E Translation ILTS By Global Matching Algorithm and POST Parser Liang Chen and Naoyuki Tokuda Computer Science Department, Utsunomiya University, Utsunomiya, Japan 321-8505 lchen@alfin.mine.utsunomiya-u.ac.jp, tokuda@cc.utsunomiya-u.ac.jp Abstract A new diagnostic system has been devel- oped for an interactive template-structured intelligent language tutoring system (ILTS) for Japanese-English translation where an efficient heaviest common sequence (HCS) matching algorithm and a ‘part-of-speech tagged (POST) parser’ play a key role. This is implemented by exploiting the system template which consists of a complex tran- sition networks comprising both model (cor- rect) translations and many typical erro- neous translations characteristic of nonna- tive beginners all collected and extracted from translations of about 200 monitors. By selecting, from among many candidates’ paths in the system template, a path hav- ing a HCS with the student’s input trans- lation as a best matched sentence, the tem- plate structure of the diagnostic system al- lows the potentially complicated bug find- ing processes in natural language to be im- plemented by a much simpler and efficient HCS string matching algorithm [20]. To improve the precision of a parser, we have developed a ‘probabilistic POST parser’ where we have eliminated ambiguity in part- of-speeches by manually pre-assigning POS tags to all words in potentially correct paths of the template. Combining the template- based diagnostic system and the parser, we found that the ILTS is capable of providing most adequate diagnostic messages and a tutoring strategy with appropriate com- ments after analyzing the keyed-in trans- lated sentences from students. 1 Introduction An Intelligent Language Tutoring System is an out- growth from extensive research results addressing the use of computers in language processing where exten- sive fruitful research results from various fields of nat- ural language processing applications have converged. While it draws heavily upon the findings of compu- tational linguistics and machine translation [11], we think that the ILTS is currently one of the most chal- lenging subject of all applications in machine transla- tion and natural language processing. While most of NLP and MT systems depend on semantic and syn- tactic analysis requiring processing well-formed sen- tences, the ILTS’ differ fundamentally from other NLP and MT applications in how and why they handle ill- formed inputs. This is essential in the ILTS appli- cations because unlike most of NLP or MT systems, the ILTS need to provide error-contingent feedback to erroneous input sentences from students for imple- menting effective tutoring. We see that the additional capability required of processing ill-formed sentences in ILTS is a basic requirement for an ITLS. Obviously it is extremely difficult to develop a syntactical parser to process them. Many ILTSs are implemented in declarative rep- resentation formalism, including notably government binding [8, 9], logical grammars [6], definite clause grammar [15], and head-driven phrase structure gram- mar [10] among others. Several ITLSs resort to Aug- mented Transition Networks [24, 22, 23, 5]. To parse ill-formed sentences, we use meta-rules by relaxing the constraints of strict grammars or …

Content truncated to 500 words