• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.18 no.6
• /
• pp.1692-1705
• /
• 2024

In this paper, we present a method that integrates a Grammar Transducer as an external language model to enhance the accuracy of the pre-trained Korean End-to-end (E2E) Automatic Speech Recognition (ASR) model. The E2E ASR model utilizes the Connectionist Temporal Classification (CTC) loss function to derive hypothesis sentences from input audio. However, this method reveals a limitation inherent in the CTC approach, as it fails to capture language information from transcript data directly. To overcome this limitation, we propose a fusion approach that combines a clause-level n-gram language model, transformed into a Weighted Finite-State Transducer (WFST), with the E2E ASR model. This approach enhances the model's accuracy and allows for domain adaptation using just additional text data, avoiding the need for further intensive training of the extensive pre-trained ASR model. This is particularly advantageous for Korean, characterized as a low-resource language, which confronts a significant challenge due to limited resources of speech data and available ASR models. Initially, we validate the efficacy of training the n-gram model at the clause-level by contrasting its inference accuracy with that of the E2E ASR model when merged with language models trained on smaller lexical units. We then demonstrate that our approach achieves enhanced domain adaptation accuracy compared to Shallow Fusion, a previously devised method for merging an external language model with an E2E ASR model without necessitating additional training.

• Journal of Internet Computing and Services
• /
• v.15 no.3
• /
• pp.125-135
• /
• 2014

In the medical field, for the purpose of diagnosis and treatment of diseases, building knowledge base has received a lot of attention. The most important thing to build a knowledge base is representing the knowledge accurately. In this paper we suggest a knowledge representation method using Ontology technique with the datasets obtained from the domestic papers on Alzheimer disease that has received a lot of attention recently in the medical field. The suggested Ontology for Alzheimer disease defines all the possible classes: lexical information from journals such as 'author' and 'publisher' research subjects extracted from 'title', 'abstract', 'keywords', and 'results'. It also included various semantic relationships between classes through the Ontology properties. Inference can be supported since our Ontology adopts hierarchical tree structure for the classes and transitional characteristics of the properties. Therefore, semantic representation based query is allowed as well as simple keyword query, which enables inference based knowledge query using an Ontology query language 'SPARQL'.

• Journal of the Korea Society of Computer and Information
• /
• v.10 no.6 s.38
• /
• pp.75-84
• /
• 2005

Bulletin board systems have evolved to support diverse multimedia data as well as text. However, current board systems have an weakness : it takes much time and efforts for users to figure out contents of articles. Most board systems provide a searching function with lexical level data access for solving that problem, however it fails to serve users' intented searching results. Moreover, it is nearly impossible to search proper articles if they contain multimedia data. This paper proposed a bulletin board system adopting the Semantic Web to solve this issue. The proposed system provides users with new ontology which is used for describing articles' domain knowledge and multimedia features. Users can describe their own board ontology using the proposed ontology. To support semantic-based searching for diverse domain knowledge without modification of the system, the system dynamically generated input/query interface and RDF data access module according to the board ontology written by administrators. The proposed board system shows that semantic-based searching is feasible and effective for users to find their intended articles.

• Language and Information
• /
• v.18 no.1
• /
• pp.33-52
• /
• 2014

This paper proposes a unified structural account of some instances of the English Modals and Semi-auxiliaries. The classification and the syntactic/structural description of the English Modal auxiliary verbs and verb-related elements have long been the center for many proposals in the history of generative syntax. According to van Gelderen (1993) and Lightfoot (2002), it was sometime around 1380 that the Tense-node (T) appeared in the phrasal structures of the English language, and the T-node is under which the English Modal auxiliaries occupy. Closely related is the existing evidence that English Modals were used as main verbs up to the early sixteenth century (Lightfoot 1991, Han 2000). This paper argues for a bi-clausal approach to English Modal auxiliaries with the infinitival particle 'to' such as 'ought to' 'used to' and 'dare (to)' 'need (to)', etc. and Semi-auxiliaries including 'be to' and 'have to'. More specifically, 'ought' in 'ought to' constructions, for instance, undergoes V-to-T movement within the matrix clause, just like 'HAVEAux' and all instances of 'BE', whereas 'to' occupies the T position of the embedded complement clause. By proposing the bi-clausal account, Radford's (2004, 2009) problems can be solved. Further, the historical motivation for the account takes a stance along with Norde (2009) and Brinton & Traugott (2005) in that Radford's (2004, 2009) syncretization of the two positions of the infinitival particle 'to' is no different from the 'boundary loss' in the process of Grammariticalization. This line of argument supports Krug's (2011), and in turn Bolinger's(1980) generalization on Auxiliaryhood, while providing a novel insight into Head movement of V-to-T in Present Day English.

• KIISE Transactions on Computing Practices
• /
• v.23 no.4
• /
• pp.226-231
• /
• 2017

Word representation, an important area in natural language processing(NLP) used machine learning, is a method that represents a word not by text but by distinguishable symbol. Existing word embedding employed a large number of corpora to ensure that words are positioned nearby within text. However corpus-based word embedding needs several corpora because of the frequency of word occurrence and increased number of words. In this paper word embedding is done using dictionary definitions and semantic relationship information(hypernyms and antonyms). Words are trained using the feature mirror model(FMM), a modified Skip-Gram(Word2Vec). Sense similar words have similar vector. Furthermore, it was possible to distinguish vectors of antonym words.

• Journal of KIISE:Software and Applications
• /
• v.29 no.10
• /
• pp.755-765
• /
• 2002

In this paper, we observe various syntactic information for Korean parsing and propose a method to learn constraints and improve the efficiency of a parsing model by using the constraints. The proposed method has the following three characteristics. First, it improves the parsing efficiency since we use constraints that can prevent the parser from generating unsuitable candidates. Second, it is robust on a given Korean sentence because the attributes for the constraints are selected based on the syntactic and lexical idiosyncrasy of Korean. Third, it is easy to acquire constraints automatically from a treebank by using a decision tree learning algorithm. The experimental results show that the parser using acquired constraints can reduce the number of overgenerated candidates up to 1/2~1/3 of candidates and it runs 2~3 times faster than the one without any constraints.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.18 no.5
• /
• pp.619-633
• /
• 1993

A new High level hardware Description Language, ASPHODEL(Algorithm Synthesis Pascal Hardware for Optimal Design and Efficient Language), and its algorithm compiler for high level synthesis are described in this paper. The new HDL, appropriated to the description of algorithmic level and lower, models VLSI circuits as an abstracted block which is consisted of input/output ports and hierachical processors to control VLSI complexities with efficiency. Also, in order to improve the descriptive power, popular Pascal programming language is modified to build ASPHODEL syntax rules. ASPHODEL algorithm compiler generates an intermediate form through lexical and syntax analysis from ASPHODEL source codes. To show the validation of presented language and its compiler, those are applied to practical design examples.

• Journal of KIISE:Software and Applications
• /
• v.35 no.5
• /
• pp.335-346
• /
• 2008

Causal relations in ontology should be defined based on the inference types necessary to solve problems specific to application as well as domain. In this paper, we present a model to define and extract causal relations for application ontology for Question-Answering (QA) on fault-diagnosis of electronic devices. Causal categories are defined by analyzing generic patterns of QA application; the relations between concepts in the corpus belonging to the causal categories are defined as causal relations. Instances of casual relations are extracted using lexical patterns in the concept definitions of domain, and extended incrementally with information from thesaurus. On the evaluation by domain specialists, our model shows precision of 92.3% in classification of relations and precision of 80.7% in identifying causal relations at the extraction phase.

• Journal of KIISE:Computing Practices and Letters
• /
• v.13 no.6
• /
• pp.427-431
• /
• 2007

In this paper we describe a Korean part-of-speech tagging approach using resolution rules for individual ambiguous word and statistical information. Our tagging approach resolves lexical ambiguities by common rules, rules for individual ambiguous word, and statistical approach. Common rules are ones for idioms and phrases of common use including phrases composed of main and auxiliary verbs. We built resolution rules for each word which has several distinct morphological analysis results to enhance tagging accuracy. Each rule may have morphemes, morphological tags, and/or word senses of not only an ambiguous word itself but also words around it. Statistical approach based on HMM is then applied for ambiguous words which are not resolved by rules. Experiment shows that the part-of-speech tagging approach has high accuracy and broad coverage.

• Journal of KIISE:Software and Applications
• /
• v.36 no.1
• /
• pp.92-108
• /
• 2009

The Princeton WordNet (PWN), which was developed during last 20 years since the mid 80, aimed at representing a mental lexicon inside the human mind. Its potentiality, applicability and portability were more appreciated in the fields of NLP and KE than in cognitive psychology. The semantic and knowledge processing is indispensable in order to obtain useful information using human languages, in the CMC and HCI environment. The PWN is able to provide such NLP-based systems with 'concrete' semantic units and their network. Referenced to the PWN, about 50 wordnets of different languages were developed during last 10 years and they enable a variety of multilingual processing applications. This paper aims at describing PWN-referenced Korean Wordnet, KorLex 1.5, which was developed from 2004 to 2007, and which contains currently about 130,000 synsets and 150,000 word senses for nouns, verbs, adjectives, adverbs, and classifiers.