We can distinguish two different groups of 'fusion expressions' depending on whether the verb (-ko) ha- 'do' can be recovered from the contracted expression concerned or not: recoverable and unrecoverable fusion expressions. Many expressions belonging to the former group show alternations between the [da]-type and the [dae]-type: e.g., o-nta-nta and o-ntae-nta '...says... is coming.' On the other hand, some expressions like ka-ntae can only be realized as the [dae]-type. The main purpose of this paper is to account for these [dae]-type expressions. First, we assume that they have an inaudible/invisible quotation $verb\;{\phi}-$, which takes as its complement a verb phrase with a neutral speech-level ending. This quotation verb is derived from the quotation verb ha-. Second, we assume that the quotation $verbs\;{\phi}-$ and ha- have the [j] sound as their stem-final element, which means that these verbs are actually represented as ${\phi}-j-$ and ha-j-, respectively. In this system, the [tae]-type expressions come out naturally from the behavior of the [j] sound. We do not employ any ad hoc phonological rules to derive the [dae]-type expressions from the corresponding [da]-type expressions.

This study proposes a Word Sense Disambiguation (WSD) algorithm, based on concept learning with special emphasis on statistically meaningful lowest frequency words. Previous works on WSD typically make use of frequency of collocation and its probability. Such probability based WSD approaches tend to ignore the lowest frequency words which could be meaningful in the context. In this paper, we show an algorithm to extract and make use of the meaningful lowest frequency words in WSD. Learning method is adopted from the Find-Specific algorithm of Mitchell (1997), according to which the search proceeds from the specific predefined hypothetical spaces to the general ones. In our model, this algorithm is used to find contexts with the most specific classifiers and then moves to the more general ones. We build up small seed data and apply those data to the relatively large test data. Following the algorithm in Yarowsky (1995), the classified test data are exhaustively included in the seed data, thus expanding the seed data. However, this might result in lots of noise in the seed data. Thus we introduce the 'maximum a posterior hypothesis' based on the Bayes' assumption to validate the noise status of the new seed data. We use the Naive Bayes Classifier and prove that the application of Find-Specific algorithm enhances the correctness of WSD.

Recently, there has been a surge of interests concerning the construction and utilization of a Korean thesaurus. In this paper, a semi-automatic method for generating a lexical-semantic network of Korean '-ha' verbs is presented through an analysis of the lexical definitions of these verbs. Initially, through the use of several tools that can filter out and coordinate lexical data, pairs constituting a word and a definition were prepared for treatment in a subsequent step. While inspecting the various definitions of each verb, we extracted and coordinated the head words from the sentences that constitute the definition of each word. These words are thought to be the main conceptual words that represent the sense of the current verb. Using these head words and related information, this paper shows that the creation of a thesaurus could be achieved without any difficulty in a semi-automatic fashion.

The linguistic characteristics of a sign language provide an important clue for an automatic construction of its expression from a given natural language sentence. For such characterization, we focus on the identification of elided constituents, the mapping of property-changing information into spatio-temporal dimension, and the need for rearranging the order of component information for enhanced quality of delivery. We use our characterization to implement a system that converts sentences in Korean into corresponding expressions in the Korean Sign Language.