Todai Robot Project


Answer the Japanese examination questions – Recognition beyond logic

“Kokugo (Nihongo)”, i.e., a Japanese examination that aims to test the ability of understanding natural language. Therefore, it requires a higher level of ability to understand natural language and is more difficult than other research themes of artificial intelligence.

Kana-kanji conversion, Word sense disambiguation

The questions which ask to write Japanese in Kanji (Chinese characters) may appear to be solved by applying Kana-Kanji (Hiragana to Kanji) conversions. In fact, Kana-Kanji conversion at present has sufficient performance in practice, however, it sometimes makes mistakes. In Kanji writing questions, different words which have the same pronunciation and are commonly confused on the test, it means that the Kana-Kanji conversion can also easily make mistakes. Therefore, in order answer to Kanji writing questions correctly, it is necessary to read around the context to get the meanings and analyze accurately.

Though it appears easy to solve word sense disambiguation questions by looking at a Japanese dictionary, it also has a difficulty similar to Kanji writing tests. These questions are simpler among the Japanese examination questions, but it is necessary to develop a method which will enable accurate judgment based on the context, with further advancement of current technology.

Reading Comprehension

Defining “What is reading comprehension” mathematically is difficult and it still has various arguments. According to analysis up until now, the first thing to note is that “logical” thinking is not tested in examinations. “Logical” here is “rational connection” in mathematical logic and formal logic (propositional logic and predicate logic, etc.) What is required to answer in an examination actually does not seem to be rational connections like above, it seems to be context or flow of sentences that is believed reasonable in human society. There is not any research that targets concepts like this clearly up until now, and it is necessary to advance analysis and research further.

Moreover, in reading a novel, it requires more complex (in other words, more illogical) processes like interpretation which reads a psychological state from the description of personage. We are pursuing analysis and data generation.

Analysis of Japanese and Chinese classics

Questions regarding Japanese and Chinese classics make up a significant portion of the university entrance examination. The type of the question is almost the same as in contemporary writings, however, analysis tools (morphological analysis, syntactic parsing) and recourses (corpus, thesaurus) for contemporary writings cannot be applied directly as they are.  Thus, it is necessary to develop tools and resources effectively and also to research methods which would enable application of tools and resources, even when those tools and resources are insufficient. Insufficiency of those resources and tools is a common problem among many languages in the world and is one of the most important themes in natural language processing.