• KIPS Transactions on Software and Data Engineering
• /
• v.7 no.4
• /
• pp.121-128
• /
• 2018

A Korean morphological analyzer adopts sequence-to-sequence (seq2seq) model, which can generate an output sequence of different length from an input. In general, a seq2seq based Korean morphological analyzer takes a syllable-unit based sequence as an input, and output a syllable-unit based sequence. Syllable-based morphological analysis has the advantage that unknown words can be easily handled, but has the disadvantages that morpheme-based information is ignored. In this paper, we propose a reranking model as a post-processor of seq2seq model that can improve the accuracy of morphological analysis. The seq2seq based morphological analyzer can generate K results by using a beam-search method. The reranking model exploits morpheme-unit embedding information as well as n-gram of morphemes in order to reorder K results. The experimental results show that the reranking model can improve 1.17% F1 score comparing with the original seq2seq model.

• Korean Journal of Cognitive Science
• /
• v.32 no.3
• /
• pp.99-115
• /
• 2021

Unlike most other alphabetic languages, letter transposition effect was not found in Korean except in the syllable level and in the morpheme level. This study was conducted in order to investigate the possible reason of the absence of letter transposition effect in Korean. Based on previous letter transposition studies, this study was to investigate on whether syllable frequency is a moderating variable and is responsible for the absence of the letter transposition effect. The results showed that significant letter transposition effect was found when a transposed non-word has high frequency syllable(e.g., 민주화 → 진무화), while such effect was not seen in a transposed non-word with low frequency syllable. The results showed that the letter transposition effect can found in Korean as well. The results also implicate the possibility that syllable frequency is the main moderating variable regarding the Korean letter transposition effect.

• English Language & Literature Teaching
• /
• v.11 no.1
• /
• pp.137-154
• /
• 2005

The purpose of this paper is to analyze vowel lengthening triggered by resyllabification. In general, short vowels become long when a following consonant is deleted, as shown in stehli 'steel' > st$\bar{e}$li and *fimf 'five' > five. We can account for the phenomenon in a straightforward way within the framework of CV phonology (Clements & Keyser, 1983) : deletion of a coda consonant C slot and then the preceding vowel spreads onto it, resulting in a long vowel. The analysis, however, cannot hold for words like ealhes 'temple', where deletion of an onset segment eventually triggers vowel lengthening in the preceding syllable. In order to account for this, I propose resyllabification. That is, ealhes first becomes eales by /h/-deletion. Next, the /l/ in coda is resyllabified as the onset of the following syllable, leaving its C slot unoccupied. Finally, the vowel spreads onto the empty slot, producing a long vowel.

• Phonetics and Speech Sciences
• /
• v.10 no.4
• /
• pp.101-110
• /
• 2018

This study aimed to show that Selkirk's concept of resyllabification offers a better analysis than Kahn's ambisyllabification to account for phonetic resyllabification. We conducted two production experiments to investigate the acoustic characteristics of the English /s/ in real words and nonce words. Ten English native speakers and six English native speakers participated in experiment 1 and experiment 2, respectively. Three acoustic cues - frication duration, center of gravity and aspiration duration of word-medial /s/ - were measured. We found that these three cues of the word-medial /s/ were realized significantly differently depending on the stresshood and openness of the preceding syllable. We preferred Selkirk's resyllabification to Kahn's ambisyllabification to explain this result because the word-medial and intervocalic /s/ behaved as the coda (as opposed to the onset) when the preceding syllable was stressed and open. The result thus suggested that two conditions must be met for the resyllabification rule to apply in English: a word-medial consonant is resyllabified only when its preceding syllable is stressed and open.

• The Journal of the Acoustical Society of Korea
• /
• v.38 no.6
• /
• pp.651-660
• /
• 2019

The present study aims to find out the necessary speech sound level which can satisfy with the speech intelligibility in a noisy classroom environments. For this, auralized materials were made to undertake listening tests with 27 people. Speech intelligibility tests were carried out using both Consonant-Vowel-Consonant (CVC) and Phonetically Balanced Words (PBW) methods. Signal to noise ratio was changed by 5 dB for each test. As a result, it was found that speech intelligibilities are increasing with larger Signal to Noise Ratio (SNR). It was also found that there is a lot of difference of speech intelligibilities by SNR for syllables (CVC) with the Reverberation Time (RT) of 1.5 s. However, any significant difference was not found for words (PBW) in the case with RTs of below 0.8 s. Also, it was revealed through the 2-way analysis of variance (ANOVA) test that SNR is the only attentive factor which can affect the Korean speech intelligibilities for both PBW and CVC methods. Therefore, RTs below 0.8 s could be the acoustic criteria for classroom which can minimize the effects of noise. In the case with RTs larger than 0.8 s, much larger SNR is needed to give sufficient speech intelligibility.