This paper presents our work on a task of automatic decomposition of text editing examples into primitive edit operations. Toward a detailed analysis of the behavior of text editing systems, identification of fine-grained edit operations performed by the systems is essential. Given a pair of source and edited sentences, the goal of our task is to generate a non-redundant sequence of primitive edit operations, i.e., the semantically minimal edit operations preserving grammaticality, that iteratively converts the source sentence to the edited sentence. First, we formalize this task, explaining its significant features and specifying the constraints that primitive edit operations should satisfy. Then, we propose a method to automate this task, which consists of two steps: generation of an edit operation lattice and selection of an optimal path. To obtain a wide range of edit operation candidates in the first step, we combine a phrase aligner and a large language model. Experimental results show that our method perfectly decomposes 44% and 64% of editing examples in the text simplification and machine translation post-editing datasets, respectively. Detailed analyses also provide insights into the difficulties of this task, suggesting directions for improvement.

This study presents an analytical evaluation of neural text simplification (TS) systems. Because recent TS models are trained in an end-to-end fashion, it is difficult to grasp their abilities to perform particular simplification operations. For the advancement of TS research and development, we should understand in detail what current TS systems can and cannot perform in comparison with human performance. To that end, we first developed an analytical evaluation framework consisting of fine-grained taxonomies of simplification strategies (at both the surface and content levels) and errors. Using this framework, we annotated TS instances produced by professional human editors and multiple neural TS systems and compared the results. Our analyses concretely and quantitatively revealed a wide gap between humans and systems, specifically indicating that systems tend to perform deletions and local substitutions while excessively omitting important information, and that the systems can hardly perform information addition operations. Based on our analyses, we also provide detailed directions to address these limitations.

Japanese sentence-ending predicates intricately combine content words and functional elements, such as aspect, modality, and honorifics; this can often hinder the understanding of language learners and children. Conventional lexical simplification methods, which replace difficult target words with simpler synonyms acquired from lexical resources in a word-by-word manner, are not always suitable for the simplification of such Japanese predicates. Given this situation, we propose a BERT-based simplification method, the core feature of which is the high ability to substitute the whole predicates with simple ones while maintaining their core meanings in the context by utilizing pre-trained masked language models. Experimental results showed that our proposed methods consistently outperformed the conventional thesaurus-based method by a wide margin. Furthermore, we investigated in detail the effectiveness of the average token embedding and dropout, and the remaining errors of our BERT-based methods.

This paper describes a coreference resolution system for math problem text. Case frame dictionaries and a math taxonomy are utilized for supplying domain knowledge. The system deals with various anaphoric phenomena beyond well-studied entity coreferences.

This paper reports the relation between text readability and word distribution in the Japanese language. There was no similar study in the past due to three major obstacles: (1) unclear definition of Japanese “word”, (2) no balanced corpus, and (3) no readability measure. Compilation of the Balanced Corpus of Contemporary Written Japanese (BCCWJ) and development of a readability predictor remove these three obstacles and enable this study. First, we have counted the frequency of each word in each text in the corpus. Then we have calculated the frequency rank of words both in the whole corpus and in each of three readability bands. Three major findings are: (1) the proportion of high-frequent words to tokens in Japanese is lower than that in English; (2) the type-coverage curve of words in the difficult-band draws an unexpected shape; (3) the size of the intersection between high-frequent words in the easy-band and these in the difficult-band is unexpectedly small.

The Japanese language has rich variety and quantity of word variant. Since 1980s, it has been recognized that this richness becomes an obstacle against natural language processing. A complete solution, however, has not been presented yet. This paper proposes a method to recognize Katakana variants―a major type of word variant in Japanese―in the process of dictionary look-up. For a given set of variant generation rules, the method executes variant generation and entry retrieval simultaneously and efficiently. We have developed the seven-layered rule set (216 rules in total) according to the specification manual of UniDic-2.1.0 and other sources. An experiment shows that the spelling-variant generator with 102 rules in the first five layers is almost perfect. Another experiment shows that the form-variant generator with all 216 rules is powerful and 77.7% of multiple spellings of Katakana loanwords are unnecessary (i.e., can be removed). This result means that the proposed method can drastically reduce the number of variants that we have to register into a dictionary in advance.

This paper proposes a simple and fast person-name filter, which plays an important role in automatic compilation of a large bilingual person-name lexicon. This filter is based on pn_score, which is the sum of two component scores, the score of the first name and that of the last name. Each score is calculated from two term sets: one is a dense set in which most of the members are person names; another is a baseline set that contains less person names. The pn_score takes one of five values, {+2, +1, 0, -1, -2}, which correspond to strong positive, positive, undecidable, negative, and strong negative, respectively. This pn_score can be easily extended to bilingual pn_score that takes one of nine values, by summing scores of two languages. Experimental results show that our method works well for monolingual person names in English and Japanese; the F-score of each language is 0.929 and 0.939, respectively. The performance of the bilingual person-name filter is better; the F-score is 0.955.

This paper describes a method of readability measurement of Japanese texts based on a newly compiled textbook corpus. The textbook corpus consists of 1,478 sample passages extracted from 127 textbooks of elementary school, junior high school, high school, and university; it is divided into thirteen grade levels and the total size is about a million characters. For a given text passage, the readability measurement method determines the grade level to which the passage is the most similar by using character-unigram models, which are constructed from the textbook corpus. Because this method does not require sentence-boundary analysis and word-boundary analysis, it is applicable to texts that include incomplete sentences and non-regular text fragments. The performance of this method, which is measured by the correlation coefficient, is considerably high (R > 0.9); in case that the length of a text passage is limited in 25 characters, the correlation coefficient is still high (R = 0.83).

This paper describes a system for finding phrasal translation correspondences from parallel parsed corpus that are collections paired English and Japanese sentences. First, the system finds phrasal correspondences by Japanese-English translation dictionary consultation. Then, the system finds correspondences in remaining phrases by using sentences dependency structures and the balance of all correspondences. The method is based on an assumption that in parallel corpus most fragments in a source sentence have corresponding fragments in a target sentence.