• Annual Conference of KIPS
• /
• 2012.04a
• /
• pp.365-368
• /
• 2012

어의 중의성 해소는 자연어처리에서 중요한 역할을 한다. 감독 중의성 해소 방법은 비감독 중의성 해소 방법보다 높은 성능을 나타내지만, 구축비용이 큰 대규모 의미부착 말뭉치가 필요하다. 본 논문에서는 중국어 어휘의미망(HowNet)과 의미 미부착 말뭉치를 이용한 중국어 비감독 어의 중의성 해소 방법을 제안한다. 의미 미부착 말뭉치에서 통계정보를 추출하고, 중국어 어휘 의미망에서 중의성 어휘의 의미별 형제어를 추출하여 중의성 어휘의 주변 문맥에 나타나는 어휘와 카이제곱검정(${\chi}^2$-test)에 의한 독립성 검정을 통해 어휘 간 연관성을 판단하고 중의성 해소를 한다. 본 논문에서 제안한 중의성 해소방법의 성능을 SemEval-2007 평가데이터에서 측정한 결과 명사와 동사에서 각각 64.7%, 49.4%를 나타냈다. 이는 SemEval-2007 중국어 비감독 중의성 해소에서 가장 높은 성능을 나타낸 시스템보다 13.1%, 13.9% 높은 성능이다.

• Annual Conference on Human and Language Technology
• /
• 2021.10a
• /
• pp.195-198
• /
• 2021

기존 연구에 따르면, 시소러스의 계층적 관계를 기반으로 압축한 의미 어휘 태그를 단어 의미 모호성 해소에 사용할 경우, 그 성능이 향상되었다. 본 논문에서는 시소러스를 사용하지 않고, 국어 사전에 포함된 단어의 의미 정의를 군집화하여 압축된 의미 어휘 태그를 만드는 방법을 제안한다. 또, 이를 이용하여 효율적으로 단어 의미 모호성을 해소하는 BERT 기반의 딥러닝 모델을 제안한다. 한국어 세종 의미 부착 말뭉치로 실험한 결과, 제안한 방법의 성능이 F1 97.21%로 기존 방법의 성능 F1 95.58%보다 1.63%p 향상되었다.

• Annual Conference on Human and Language Technology
• /
• 2022.10a
• /
• pp.268-271
• /
• 2022

단어 의미 모호성 해소는 동형이의어의 의미를 문맥에 맞게 결정하는 일이다. 최근 연구에서는 희소 데이터 처리를 위해 시소러스를 사용해 의미 어휘를 압축하고 사용하는 방법이 좋은 성능을 보였다[1]. 본 연구에서는 시소러스 없이 군집화 알고리즘으로 의미 어휘를 압축하는 방법의 성능 향상을 위해 두 가지 방법을 제안한다. 첫째, 의미적으로 유사한 의미 어휘 집합인 범주(category) 정보를 군집화를 위한 초기 군집 생성에 사용한다. 둘째, 다양하고 많은 문맥 정보를 학습해 만들어진 품질 좋은 벡터를 군집화에 사용한다. 영어데이터인 SemCor 데이터를 학습하고 Senseval, Semeval 5개 데이터로 평가한 결과, 제안한 방법의 평균 성능이 기존 연구보다 1.5%p 높은 F1 70.6%를 달성했다.

• Journal of KIISE:Software and Applications
• /
• v.27 no.7
• /
• pp.769-783
• /
• 2000

In this paper, in order to disambiguate Korean noun word sense, we define a local context and explain how to extract it from a raw corpus. Following the intuition that two different nouns are likely to have similar meanings if they occur in the same local context, we use, as a clue, the word that occurs in the same local context where the target noun occurs. This method increases the usability of extracted knowledge and makes it possible to disambiguate the sense of infrequent words. And we can overcome the data sparseness problem by extending the verbs in a local context. The sense of a target noun is decided by the maximum similarity to the clues learned previously. The similarity between two words is computed by their concept distance in the sense hierarchy borrowed from WordNet. By reducing the multiplicity of clues gradually in the process of computing maximum similarity, we can speed up for next time calculation. When a target noun has more than two local contexts, we assign a weight according to the type of each local context to implement the differences according to the strength of semantic restriction of local contexts. As another knowledge source, we get a co-occurrence information from dictionary definitions and example sentences about the target noun. This is used to support local contexts and helps to select the most appropriate sense of the target noun. Through experiments using the proposed method, we discovered that the applicability of local contexts is very high and the co-occurrence information can supplement the local context for the precision. In spite of the high multiplicity of the target nouns used in our experiments, we can achieve higher performance (89.8%) than the supervised methods which use a sense-tagged corpus.

• Journal of the Korean Society for Library and Information Science
• /
• v.45 no.2
• /
• pp.251-275
• /
• 2011

In this paper, we propose a novel kernel called a semantic parse tree kernel that extends the parse tree kernel previously studied to extract protein-protein interactions(PPIs) and shown prominent results. Among the drawbacks of the existing parse tree kernel is that it could degenerate the overall performance of PPI extraction because the kernel function may produce lower kernel values of two sentences than the actual analogy between them due to the simple comparison mechanisms handling only the superficial aspects of the constituting words. The new kernel can compute the lexical semantic similarity as well as the syntactic analogy between two parse trees of target sentences. In order to calculate the lexical semantic similarity, it incorporates context-based word sense disambiguation producing synsets in WordNet as its outputs, which, in turn, can be transformed into more general ones. In experiments, we introduced two new parameters: tree kernel decay factors, and degrees of abstracting lexical concepts which can accelerate the optimization of PPI extraction performance in addition to the conventional SVM's regularization factor. Through these multi-strategic experiments, we confirmed the pivotal role of the newly applied parameters. Additionally, the experimental results showed that semantic parse tree kernel is superior to the conventional kernels especially in the PPI classification tasks.

• Journal of Korea Multimedia Society
• /
• v.25 no.3
• /
• pp.493-501
• /
• 2022

This paper is a study on context-sensitive spelling error correction and uses the Korean WordNet (KorLex)[1] that defines the relationship between words as a graph to improve the performance of the correction[2] based on the vector information of the word embedded in the correction technique. The Korean WordNet replaced WordNet[3] developed at Princeton University in the United States and was additionally constructed for Korean. In order to learn a semantic network in graph form or to use it for learned vector information, it is necessary to transform it into a vector form by embedding learning. For transformation, we list the nodes (limited number) in a line format like a sentence in a graph in the form of a network before the training input. One of the learning techniques that use this strategy is Deepwalk[4]. DeepWalk is used to learn graphs between words in the Korean WordNet. The graph embedding information is used in concatenation with the word vector information of the learned language model for correction, and the final correction word is determined by the cosine distance value between the vectors. In this paper, In order to test whether the information of graph embedding affects the improvement of the performance of context- sensitive spelling error correction, a confused word pair was constructed and tested from the perspective of Word Sense Disambiguation(WSD). In the experimental results, the average correction performance of all confused word pairs was improved by 2.24% compared to the baseline correction performance.

• Proceedings of the IEEK Conference
• /
• summer
• /
• pp.103-109
• /
• 2004

In NLP (Natural Language Processing) tasks, the highest difficulty which computers had to face with, is the built-in ambiguity of Natural Languages. To disambiguate it, formerly, they based on human-devised rules. Building such a complete rule-set is time-consuming and labor-intensive task whilst it doesn't cover all the cases. Besides, when the scale of system increases, it is very difficult to control that rule-set. So, recently, many NLP tasks have changed from rule-based approaches into corpus-based approaches with large annotated corpora. Corpus-based NLP tasks for such popular languages as English, French, etc. have been well studied with satisfactory achievements. In contrast, corpus-based NLP tasks for Vietnamese are at a deadlock due to absence of annotated training data. Furthermore, hand-annotation of even reasonably well-determined features such as part-of-speech (POS) tags has proved to be labor intensive and costly. In this paper, we present our building an annotated English-Vietnamese parallel aligned corpus named EVC to train for Vietnamese-related NLP tasks such as Word Segmentation, POS-tagger, Word Order transfer, Word Sense Disambiguation, English-to-Vietnamese Machine Translation, etc.

• Annual Conference on Human and Language Technology
• /
• 2019.10a
• /
• pp.272-275
• /
• 2019

단어 중의성 해소 방법은 지식 정보를 활용하여 문제를 해결하는 지식 기반 방법과 각종 기계학습 모델을 이용하여 문제를 해결하는 지도학습 방법이 있다. 지도학습 방법은 높은 성능을 보이지만 대량의 정제된 학습 데이터가 필요하다. 반대로 지식 기반 방법은 대량의 정제된 학습데이터는 필요없지만 높은 성능을 기대할수 없다. 최근에는 이러한 문제를 보완하기 위해 지식내에 있는 정보와 정제된 학습데이터를 기계학습 모델에 학습하여 단어 중의성 해소 방법을 해결하고 있다. 가장 많이 활용하고 있는 지식 정보는 상위어(Hypernym)와 하위어(Hyponym), 동의어(Synonym)가 가지는 의미설명(Gloss)정보이다. 이 정보의 표상을 기존의 문장의 표상과 같이 활용하여 중의성 단어가 가지는 의미를 파악한다. 하지만 정확한 문장의 표상을 얻기 위해서는 단어의 표상을 잘 만들어줘야 하는데 기존의 방법론들은 모두 문장내의 문맥정보만을 파악하여 표현하였기 때문에 정확한 의미를 반영하는데 한계가 있었다. 본 논문에서는 의미정보와 문맥정보를 담은 단어의 표상정보를 만들기 위해 구문정보, 의미관계 그래프정보를 GCN(Graph Convolutional Network)를 활용하여 임베딩을 표현하였고, 기존의 모델에 반영하여 문맥정보만을 활용한 단어 표상보다 높은 성능을 보였다.

• Journal of the Korea Computer Industry Society
• /
• v.2 no.12
• /
• pp.1581-1590
• /
• 2001

This paper asserted more developmental model of WSD(word sense disambiguation) than J. Hur(2000)'s WSD model. This model suggested an improved statistical homonym disambiguation Model based on Bayes Theory. This paper using semantic information(co-occurrence data) obtained from definitions of part of speech(POS) tagged UMRD-S(Ulsan university Machine Readable Dictionary(Semantic Tagged)). we extracted semantic features in the context as nouns, predicates and adverbs from the definitions in the korean dictionary. In this research, we make an experiment with the accuracy of WSD system about major nine homonym nouns and new seven homonym predicates supplementary. The inner experimental result showed average accuracy of 98.32% with regard to the most Nine homonym nouns and 99.53% for the Seven homonym predicates. An Addition, we save test on Korean Information Base and ETRI's POS tagged corpus. This external experimental result showed average accuracy of 84.42% with regard to the most Nine nouns over unsupervised learning sentences from Korean Information Base and ETRI Corpus, 70.81 ％ accuracy rate for the Seven predicates from Sejong Project phrase part tagging corpus (3.5 million phrases) too.

• Journal of KIISE
• /
• v.43 no.1
• /
• pp.71-79
• /
• 2016

Disambiguation of homographs is an important job in Korean semantic processing and has been researched for long time. Recently, machine learning approaches have demonstrated good results in accuracy and speed. Other knowledge-based approaches are being researched for untrained words. This paper proposes a hybrid method based on the machine learning approach that uses a lexical semantic network. The use of a hybrid approach creates an additional corpus from subcategorization information and trains this additional corpus. A homograph tagging phase uses the hypernym of the homograph and an additional corpus. Experimentation with the Sejong Corpus and UWordMap demonstrates the hybrid method is to be effective with an increase in accuracy from 96.51% to 96.52%.