This paper presents a model to analyze the coherence relation within complex sentences by using the attributes of verbs and subjects.The relations between subordinate and main clauses can not be understood only by the connectives that link them. The coherence relations by connectives are often ambiguous. For example, a Japanese conjunctive particle “te” expresses coherence relations, such as sequential, method, manner, reason, purpose and parallel. The coherence relations by connectives depend on the semantic types of the predicates in subordinate and main clauses, and the combination of them. It is presumed that there are the patterns of coherence relations in the relations between subordinate and main clauses, just like there are the patterns of thematic roles to express semantic relations between verbs and nouns. The pattern of coherence relation uses the attributes of verbs and subjects in a subordinate and a main clause, just like the pattern of thematic roles uses a verb and the semantic types of nouns. The model infers coherence relations by using the attributes of verbs and subjects in subordinate and main clauses. It uses volition, semantic types, idiomatic expressions, mood/aspect/voice, as the attributes of verbs, and uses whether the subject in a subordinate clause is identical with the one in the main clause or not, whether the subject is unanimate or not as the attributes of subjects. The model is evaluated and the result shows that 95% of the text taken from science and technical documents can be analyzed successfully.

View full abstract

This paper proposes a new method for ordering English adverbs.First, we propose a classification of adverbs for English adverb generation.Adverbs are classified into 41 classes by grammatical function (adjuncts, subjuncts, disjuncts and conjuncts), meaning (process, space, time etc.) and their default positions in sentences (initial, medial, end, pre, and post). Then a method to order English adverbs correctly is described, using the proposed adverb classification and principles of word ordering for adverbs (principles of ordering between adverbs and other sentence constituents and principles of ordering between adverbs). In particular we give detailed rules for deciding precedence when two adverbs have the same default position. Exceptions to the default adverb generating process are also described. Finally, the proposed method is examined in three experiments from the point of view of Japanese-to-English machine translation. The first experiment focuses on aspects of various types of adverbs and a comparison of the proposed method and the previous method. The second experiment focuses on aspects of quantitative coverage, and the third looks at aspects of practical use. The results show an accuracy of 97% or more in all experiments which highlights the efficiency of the proposed method. The third experiment, in particular, with an accuracy of 99%, confirms that the proposed method is effective in practical applications.

View full abstract

The study on aspect in Japanese has evolved from the description of the meaning for each type such as “progressive” or “perfective” into the process of the determination of the aspectual meaning including adverbial phrases etc. In this paper, we consider the meaning of the aspectual markers or adverbs as the indicator of the cognitive process like “zooming” or “focusing” on the time-line and define them as the dynamic operations on the meaning of the verb phrases.We estimate the aspectual meaning of verbs from surface expressions in a corpus and represent them as a bundle of features.To evaluate the result of the experiment, we examined the meaning of si-teiru which is one of the most fundamental aspectual markers, and obtained the correct recognition score of 71% for the 200 sentences.

View full abstract

We discuss here how a discourse structure for a causal relation between two or more utterances is linguistically expressed in Japanese task-free conversations. When we have a conversation, we often use conjunctions as a sign of causal relation, change of topic, etc. So, these words are important to understand a discourse structure because we often use these words to represent “an intention of utterance” or “a stream of discourse.” Here, we discuss a case where causal conjunctions are used in Japanese task-free conversations by examining our corpus of causal conversation, and show how to identify relations, represented by these conjunctions, among a few adjacent sentences systematically.

View full abstract

One of the most interesting issues in corpus-based studies is deriving linguistic knowledge via automated procedures. Most works, however, have focused on deriving lexico-syntactic knowledge. In the work described here, we automatically deduce dialogue models from a corpus with probabilistic methods. The corpus is a subset of the ATR Dialogue Database, and consists of simulated dialogues between a secretary and a questioner at international conferences. Each utterance is annotated with a speaker label and an utterance type, called IFT (Illocutionary Force Type), which is an abstraction of the speaker's intention in terms of the type of action the speaker intends by the utterance. We use two kinds of probabilistic methods to model the speaker-IFT sequences of the corpus: (1) an Ergodic HMM (Hidden Markov Model) and (2) the ALERGIA algorithm, an algorithm for learning probabilistic automata by means of state merging. By analyzing the derived dialogue models, we see that both methods capture the basic characteristics of the local discourse structure, such as turn-taking and speech act sequencing. We also describe the quality measurement of the dialogue models from the information-theoretic viewpoint.

View full abstract