• Journal of KIISE:Software and Applications
• /
• v.34 no.2
• /
• pp.112-122
• /
• 2007

Training a statistical model for semantic role labeling requires a large amount of manually tagged corpus. However. such corpus does not exist for Korean and constructing one from scratch is a very long and tedious job. This paper suggests a modified algorithm of self-training, an unsupervised algorithm, which trains a semantic role labeling model from any raw corpora. For initial training, a small tagged corpus is automatically constructed iron case frames in Sejong Electronic Dictionary. Using the corpus, a probabilistic model is trained incrementally, which achieves 83.00% of accuracy in 4 selected adverbial cases.

• Language and Information
• /
• v.9 no.2
• /
• pp.85-105
• /
• 2005

This study aims at investigating the 'verb-second' phenomena indicated in complex sentences of medieval Spanish. Especially, when the complex sentence is composed of a preposed adverbial clause and its succeeding main clause, the subject inversion is noticeable in the latter. The fundamental motive of this type of inversion is due to the 'verb-second' structure, in which a topic appears in the first position and the verb immediately after the topic. So it can be said that the subject inversion is a prerequisite for a verb to be located in the second position when the adverbial clause functions as a topic to the main clause, as is often the case with Germanic languages like German, Dutch, etc.. On the contrary, modern Spanish complex sentences do not show this phenomenon, with a strong tendency to locate a grammatical subject in the preverbal position. Therefore, medieval Spanish might be typologically closer to Germanic languages than to modern Spanish. In order to argue for this assumption, the formal and functional criteria by which the preposed adverbial clause could be defined as a topic NP will be examined across the comparition with left-dislocation structure.

• Journal of KIISE:Software and Applications
• /
• v.27 no.6
• /
• pp.668-677
• /
• 2000

Korean has the characteristics that case postpositions determine the syntactic roles of phrases and a postposition may have more than one meanings. In particular, the adverbial postpositions make translation from Korean to English difficult, because they can have various meanings. In this paper, we describe a method for resolving such semantic ambiguities of Korean adverbial postpositions using decision trees. The training examples for decision tree induction are extracted from a corpus consisting of 0.5 million words, and the semantic roles for adverbial postpositions are classified into 25 classes. The lack of training examples in decision tree induction is overcome by clustering words into classes using a greedy clustering algorithm. The cross validation results show that the presented method achieved 76.2% of precision on the average, which means 26.0% improvement over the method determining the semantic role of an adverbial postposition as the most frequently appearing role.

• Language and Information
• /
• v.6 no.2
• /
• pp.1-14
• /
• 2002

This paper discusses so called the non-DP subject constructions in English. In general, a subject is a DP that bears Nominative case and that occupies 〔Spec, IP〕. However, in some examples under investigation, it looks as if non-DP categories such as Prepositional Phrases(PP), Adjectival Phrases(AP), Adverbial Phrases (AdvP), Small Clauses (PreP or SC), and VP occupy the canonical subject position, 〔Spec, IP〕. Under the framework of Chomsky's (1993, 1995) along with his previous works (Chomsky 1981, 1986), the Case Checking mechanism undoubtedly assumes that only DPs can have Case Therefore, the Case Checking/Agree mechanism is stated such that the strong uninterpretable feature, in this case Case feature (D or NP) feature must be checked off in a certain manner. Therefore, any phrasal categories other than DPs are not included in the considerations. Nonetheless, there are many instances of non-DP categories in English that occupy the seemingly canonical subject position, 〔spec, IP〕. In this paper, it is proposed that the actual position of these non-DP subjects in English is not in Spec of IP. Rather, they occupy 〔Spec, TopP〕 under CP in the sense of Lasnik & Stowell (1991), Rizzi (1997), and Haegeman & Gueron (1999). In its effect, therefore, this paper extends the idea of Stowell (1981) who argues that the clausal subjects in English is not in 〔Spec, IP〕, but in 〔Spec, TopP〕. We further argue that Stowell's version of Case Resistance Principle must be extended in order to accomodate many more occurrences of so called non-DP subjects.

• Korean Journal of Child Studies
• /
• v.29 no.6
• /
• pp.147-161
• /
• 2008

The present study compared characteristics of Korean nominal particles (occurrence, error, error patterns) of ten 5- to 6-year-old Korean-English simultaneous bilingual children with ten Korean monolingual children. Data were analyzed by Mann-Whitney U test and Spearman Rank Correlation and by qualitative analysis. Results were (1) bilingual children showed significantly lower frequency based on the number of occurrence of nominal particles per utterance. (2) The error percentage of adverbial markers was significantly higher for bilingual children. (3) Error patterns of bilingual children showed a higher percentage of in-case substitution and double use error. These findings suggest that Korean-English simultaneous bilingual children have a different Korean nominal particles development from Korean monolingual children.

• Journal of KIISE:Software and Applications
• /
• v.32 no.10
• /
• pp.984-989
• /
• 2005

This study aims to identify grammatical relations (GRs) in Korean sentences. The key task is to find the GRs in sentences in terms of such GR categories as subject, object, and adverbial. To overcome this problem, we are fared with the many ambiguities. We propose a statistical model, which resolves the grammatical relational ambiguity first, and then finds correct noun phrases (NPs) arguments of given verb phrases (VP) by using the probabilities of the GRs given NPs and VPs in sentences. The proposed model uses the characteristics of the Korean language such as distance, no-crossing and case property. We attempt to estimate the probabilities of GR given an NP and a VP with Support Vector Machines (SVM) classifiers. Through an experiment with a tree and GR tagged corpus for training the model, we achieved an overall accuracy of $84.8\%,\;94.1\%,\;and\;84.8\%$ in identifying subject, object, and adverbial relations in sentences, respectively.

• Annual Conference on Human and Language Technology
• /
• 2005.10a
• /
• pp.120-126
• /
• 2005

세종전자사전의 용언사전과 체언사전에 기재된 용언의 격틀과 명사의 의미부류는 문장의 의미분석을 위한 핵심적인 언어자원이다. 본 논문에서는 용언사전을 전산처리가 용이한 격틀사전으로 변형한 다음 이를 이용한 의미역 결정 시스템을 구축하였고 기계학습 방법에 기반한 의미역 결정 시스템과 혼합하여 한국어에 있어 '에, 로'를 격표지로 하는 부사격에 대한 의미역 결정 방법에 대해 다루고 있다.

• Proceedings of the Korean Information Science Society Conference
• /
• 2005.07b
• /
• pp.451-453
• /
• 2005

본 논문에서는 한국어에 있어 '에, 로'를 격표지로 하는 부사격에 대한 의미역 결정 모델에 대해 다루고 있다. 의미역 결정은 의미 분석의 핵심 과정 중 하나이고 자연언어처리에서 해결해야 할 중요한 문제이다. 본 논문은 기존 연구와 언어학 논저를 참고해서 의미역 결정에 유용한 자질들을 정리하였고 SVM을 이용하여 의미역 결정 모델을 구축하였다. 또한 기존 연구와 차별적으로 기능동사 구문의 처리와 지배소 개념의 유사도 보정 방법을 사용하여 보다 견고한 모델을 만들 수 있었다. 성능 평가 결과 개념(Concept)만을 사용한 기본 모델에 비해서 평균 $9\%$의 정확률 향상을 보였다.

• Annual Conference on Human and Language Technology
• /
• 2006.10e
• /
• pp.32-39
• /
• 2006

본 논문은 한국어정보처리 과정에서 구문 관계를 의미 관계로 사상하는 의미역 결정 문제에 대해 다루고 있다. 한국어의 경우 대량의 학습 말뭉치를 구하기 힘들며, 이를 구축하기 위해서는 많은 시간과 노력이 필요한 문제점이 있다. 따라서 본 논문에서는 학습 말뭉치를 직접 태깅하지 않고 격틀사전을 이용하여 자동으로 학습 말뭉치를 구축하고 간단한 확률모델을 적용하여 점진적으로 모델을 학습하는 수정된 self-training 알고리즘을 사용하였다. 실험 결과, 4개의 부사격 조사에 대해 평균적으로 81.81%의 정확률을 보였으며, 수정된 self-training 방법은 기존의 방법에 비해 성능 및 실행시간에서 개선된 결과를 보였다.

• The KIPS Transactions:PartB
• /
• v.15B no.1
• /
• pp.69-72
• /
• 2008

This study aims to identify dependency structures in Korean sentences with the cascaded chunking. In the first stage of the cascade, we find chunks of NP and guess grammatical relations (GRs) using Support Vector Machine (SVM) classifiers for all possible modifier-head pairs of chunks in terms of GR categories as subject, object, complement, adverbial, etc. In the next stages, we filter out incorrect modifier-head relations in each cascade for its corresponding GR using the SVM classifiers and the characteristics of the Korean language such as distance between relations, no-crossing and case property. Through an experiment with a parsed and GR tagged corpus for training the proposed parser, we achieved an overall accuracy of 85.7%.