Research on the Cognitive Process of second language vocabulary learning and its application to machine translation

我们调查了成人学习者在学习第二语言时面临的困难的原因，特别关注日本人学习英语作为第二语言。具体来说，我们怀疑第二语言学习困难的原因之一是由于学习者的母语和他/她正在学习的目标第二语言之间的词汇化结构的差异。我们以两个语义领域为重点对这一假设进行了实证检验：(1)词汇化运动领域，以及(2)个性化方面的词汇化上级概念领域。通过一系列的实验，我们发现，在这两个领域中，日语使用者都将新英语单词假设为对应的日语单词，并且这种假设阻碍了目标英语单词的完整含义的学习。为实现高质量的机器翻译进行了基础研究。对小学教科书中抽取的日语基本名词进行了基因实现概念、特定概念、相似概念、属性概念、相关动作概念和相关情景概念的收集实验。这些概念被组织成层次结构，形成一个电子概念词典。一般来说，上下文信息对于高质量的机器翻译是必要的。词典将从电子概念词典中的相关动作概念或相关情境概念中提供这些关键信息。对于人类词汇学习机制的计算机模拟，我们尝试用溯因性归纳逻辑规划（溯因性ILP）来表述。溯因性ILP不同于标准的ILP，在某种意义上，它更倾向于增加背景知识，而不是推断规则来解释给定的例子。通过使用这个功能，系统可以猜测，例如，在给定的句子中，由于缺乏信息而无法正确解析的某些单词的语法类别缺失。在我们的研究中，我们开发了一套完整的程序来寻找缺失的单个规则或事实，如果存在的话。少

We irrvestigated the causes for difficuties adult learners face in learning a second language, with a special focus on Japanese speakers learning English as the second language. Specifically, we suspected that one of the reasons for the difficulty in learning a second language is due to the difference in lexicalzation structures between the learner's narive language and the target second language he/she is learning We tested this hypothesis empirically focusing on the two semantic domains : (1) a domain of lexicalizing motions, and (2) the domain of lexicalizing superordinate concepts with respect to individuation. Through a series of experiments, we found that, in both domains, Japanese speakers assumed novel English words as equivalents of the corresponding Japanese words, and learning of the full meanings of the target English words were hindered by this assumption.The basic research for implementing the high quality machine translation has been conducted. A large scale association … More experiment was implemented to collect gene ralized concepts, specific ones, similar ones, attribute ones, related action concepts and related situation ones from Japanese basic nouns which are extracted from elementary school text books. The concepts are organized into hierarchical structures to be an electronic concept dictionary. In general, contextual information is necessary for high quality machine translation. The dictionary will provide such crucial information from the related action concepts or the related situation concepts in the electronic concept dictionary.As for computer simulation of human vocaburary learning mechanism, we tried to formulate it by Inductive Logic Programming with abduction (abductive ILP). Abductive ILP is different from standard ILP in a sense that it rather suggests to augment background knowledge than to infer rules to explain given examples. By using this function, the system can guess, for example, a missing grammatical category for certain words in a given sentence when it cannot be parsed correctly due to the lacking information. In our research, we developed a complete procedure to find the missing a single rule or fact if there exist. Less