This paper proposes a new computational method for comprehending attributional metaphors. The proposed method generates deeper interpretations of metaphors than other methods through the process of figurative mapping that transfers affectively similar features of the source concept onto the target concept. Any features are placed on a common two-dimensional space revealed in the domain of psychology, and similarity of two features is calculated as a distance between them in the space. A computational model of metaphor comprehension based on the method has been implemented in a computer program called PROMIME (PROtotype system of Metaphor Interpreter with MEtaphorical mapping). Comparison between the PROMIME system's output and human interpretation shows that the performance of the proposed method is satisfactory.

抄録全体を表示

This paper presents a new formalization of probabilistic GLR (PGLR) language modeling for statistical parsing.Our model inherits its essential features from Briscoe and Carroll's generalized probabilistic LR model (Briscoe and Carroll 1993), which takes context of parse derivation into account by assigning a probability to each LR parsing action according to its left and right context. Briscoe and Carroll's model, however, has a drawback in that it is not formalized in any probabilistically well-founded way, which may degrade its parsing performance. Our formulation overcomes this drawback with a few significant refinements, while maintaining all the advantages of Briscoe and Carroll's modeling. In this paper, we discuss the formal and qualitative aspects of our PGLR model, illustrating the qualitative differences between Briscoe and Carroll's model and our model, and their expected impact on parsing performance.

抄録全体を表示

This paper proposes a new class-based method for estimating the strength of association for word co-occurrences. To deal with sparseness of data we first use a conceptual dictionary to find upper classes of the words in the co-occurrence relation. We then use t-scores to determine a pair of classes to be used in the estimation. The strength of association for the word co-occurrence is calculated using the classes thus obtained. We have applied our method to determining dependency relations in Japanese and prepositional phrase attachment in English. The experimental results show that our method is sound, effective and useful in resolving different types of ambiguities.

抄録全体を表示

長い文の統語解析を行なうと, 多くの解析結果が得られることがある. これは, その文に含まれる曖昧性に起因するものであり, 曖昧性が大きくなるにつれて, 解析結果の数が増え, 解析時間も長くなる. 曖昧性を解消する手法は古くから提案されているが, 近年は, 統計的に学習した優先度情報を文法制約に組み込んで, 解析結果問の優劣を判断し, 曖昧性を解消する手法が広く用いられている. しかし, 統計的に得られた優先度情報を文法制約に組み込むことは, 知識のモジュール性の観点から, あまり望ましいことではない. 本論文では, 文法制約と統計的に得られた優先度情報を, 知識レベルでは独立に記述しておき, 統語解析を行なう処理レベルで, 統合して利用する手法を提案する. 優先度情報としては, 統計的に学習した確率付き係り受け情報を利用する. 優先度の高い係り受け関係から順に成立させることができるように, チャート法のアルゴリズムに修正を加えた. その結果, 統計情報と文法制約を処理レベルで統合的に利用することができ, また, 解析の効率化も実現できる. 講談社和英辞典の例文220文について実験を行なったところ, 解析中に生成される弧の数が4分の1, 解析速度が7倍になった.

抄録全体を表示

本論文では, 構文解析の曖昧性解消を行うために, 構文的な統計情報と語彙的な統計情報を統合する手法を提案する. 我々が提案する統合的確率言語モデルは, 構文的優先度などの構文的な統計情報を反映する構文モデルと, 単語の出現頻度や単語の共起関係などの語彙的な統計情報を反映する語彙モデルの2つの下位モデルから成る. この統合的確率言語モデルは, 構文的な統計情報と語彙的な統計情報を同時に学習する過去の多くのモデルと異なり, 両者を個別に学習する点に特徴がある. 構文的な統計情報と語彙的な統計情報を独立に取り扱うことにより, それぞれの統計情報を異なる言語資源から独立に学習することができるだけでなく, それぞれの統計情報が曖昧性解消においてどのような効果を果たすのかを容易に分析することができる. この統合的確率言語モデルを評価するために, 日本語文の文節の係り受け解析を行った. 構文モデルを用いたときの文節の正解率は73.38%となり, ベースラインに比べて11.70%向上した. また, 構文モデルと語彙モデルを組み合わせることにより, 文節の正解率はさらに10.96%向上し84.34%となった. この結果, 本研究で提案する枠組において, 語彙的な統計情報は構文的な統計情報と同程度に曖昧性解消に貢献することを確認した.

抄録全体を表示

This paper presents a method for inducing a context-sensitive conditional probability context-free grammar from an unlabeled bracketed corpus using local contextual information and describes a natural language parsing model which uses a probabilitybased scoring function of the grammar to rank parses of a sentence. This method uses clustering techniques to group brackets in a corpus into a number of similar bracket groups based on their local contextual information. From the set of these groups, the corpus is automatically labeled with some nonterminal labels, and consequently a grammar with conditional probabilities is acquired. Based on these conditional probabilities, the statistical parsing model provides a framework for finding the most likely parse of a sentence. A number of experiments are made using EDR corpus and Wall Street Journal corpus. The results show that our approach achieves a relatively high accuracy: 88% recall, 72% precision and 0.7 crossing brackets per sentence for sentences shorter than 10 words, and 71% recall, 51% precision and 3.4 crossing brackets for sentences between 10-19 words. This result supports the assumption that local contextual statistics obtained from an unlabeled bracketed corpus are effective for learning a useful grammar and parsing.

抄録全体を表示