• Journal of the Korea Society of Computer and Information
• /
• v.24 no.11
• /
• pp.41-49
• /
• 2019

We are to build an unsupervised machine learning-based language model which can estimate the amount of information that are in need to process words consisting of subword-level morphemes and syllables. We are then to investigate whether the reading times of words reflecting their morphemic and syllabic structures are predicted by an information-theoretic measure such as surprisal. Specifically, the proposed Morfessor-based unsupervised machine learning model is first to be trained on the large dataset of sentences on Sejong Corpus and is then to be applied to estimate the information-theoretic measure on each word in the test data of Korean words. The reading times of the words in the test data are to be recruited from Korean Lexicon Project (KLP) Database. A comparison between the information-theoretic measures of the words in point and the corresponding reading times by using a linear mixed effect model reveals a reliable correlation between surprisal and reading time. We conclude that surprisal is positively related to the processing effort (i.e. reading time), confirming the surprisal hypothesis.

• Smart Media Journal
• /
• v.7 no.4
• /
• pp.17-23
• /
• 2018

We proposed a LSTM-based RNN model that can effectively perform the automatic spacing characteristics. For those long or noisy sentences which are known to be difficult to handle within Neural Network Learning, we defined a proper input data format and decoding data format, and added dropout, bidirectional multi-layer LSTM, layer normalization, and attention mechanism to improve the performance. Despite of the fact that Sejong corpus contains some spacing errors, a noise-robust learning model developed in this study with no overfitting through a dropout method helped training and returned meaningful results of Korean word spacing and its patterns. The experimental results showed that the performance of LSTM sequence-to-sequence model is 0.94 in F1-measure, which is better than the rule-based deep-learning method of GRU-CRF.

• The KIPS Transactions:PartB
• /
• v.12B no.7 s.103
• /
• pp.829-840
• /
• 2005

Word Sense Disambiguation(WSD) is one of the most difficult problem in Korean information processing. We propose a WSD model with the capability to filter semantic information using the specific characteristics in dictionary dictions, and nth added information, useful to sense determination, such as statistical, distance and case information. we propose a model, which can resolve the issues resulting from the scarcity of semantic information data based on the word hierarchy system (thesaurus) developed by Ulsan University's UOU Word Intelligent Network, a dictionary-based toxicological database. Among the WSD models elaborated by this study, the one using statistical information, distance and case information along with the thesaurus (hereinafter referred to as 'SDJ-X model') performed the best. In an experiment conducted on the sense-tagged corpus consisting of 1,500,000 eojeols, provided by the Sejong project, the SDJ-X model recorded improvements over the maximum frequency word sense determination (maximum frequency determination, MFC, accuracy baseline) of $18.87\%$ ($21.73\%$ for nouns and inter-eojeot distance weights by $10.49\%$ ($8.84\%$ for nouns, $11.51\%$ for verbs). Finally, the accuracy level of the SDJ-X model was higher than that recorded by the model using only statistical information, distance and case information, without the thesaurus by a margin of $6.12\%$ ($5.29\%$ for nouns, $6.64\%$ for verbs).

• KIPS Transactions on Software and Data Engineering
• /
• v.11 no.7
• /
• pp.291-298
• /
• 2022

In this paper, we present a method for dependency parsing with chunking in Korean. Dependency parsing is a task of determining a governor of every word in a sentence. In general, we used to determine the syntactic governor in Korean and should transform the syntactic structure into semantic structure for further processing like semantic analysis in natural language processing. There is a notorious problem to determine whether syntactic or semantic governor. For example, the syntactic governor of the word "먹고 (eat)" in the sentence "밥을 먹고 싶다 (would like to eat)" is "싶다 (would like to)", which is an auxiliary verb and therefore can not be a semantic governor. In order to mitigate this somewhat, we propose a Korean dependency parsing after chunking, which is a process of segmenting a sentence into constituents. A constituent is a word or a group of words that function as a single unit within a dependency structure and is called a chunk in this paper. Compared to traditional dependency parsing, there are some advantage of the proposed method: (1) The number of input units in parsing can be reduced and then the parsing speed could be faster. (2) The effectiveness of parsing can be improved by considering the relation between two head words in chunks. Through experiments for Sejong dependency corpus, we have shown that the USA and LAS of the proposed method are 86.48% and 84.56%, respectively and the number of input units is reduced by about 22%p.