한글 음소 단위 딥러닝 모형을 이용한 감성분석

이재준,권순범,안성만 한국IT서비스학회 2018 한국IT서비스학회지 Vol.17 No.1

Sentiment analysis is a technique of text mining that extracts feelings of the person who wrote the sentence like movie review. The preliminary researches of sentiment analysis identify sentiments by using the dictionary which contains negative and positive words collected in advance. As researches on deep learning are actively carried out, sentiment analysis using deep learning model with morpheme or word unit has been done. However, this model has disadvantages in that the word dictionary varies according to the domain and the number of morphemes or words gets relatively larger than that of phonemes. Therefore, the size of the dictionary becomes large and the complexity of the model increases accordingly. We construct a sentiment analysis model using recurrent neural network by dividing input data into phoneme-level which is smaller than morpheme-level. To verify the performance, we use 30,000 movie reviews from the Korean biggest portal, Naver. Morpheme-level sentiment analysis model is also implemented and compared. As a result, the phoneme- level sentiment analysis model is superior to that of the morpheme-level, and in particular, the phoneme-level model using LSTM performs better than that of using GRU model. It is expected that Korean text processing based on a phoneme-level model can be applied to various text mining and language models.

L2 습득이론에 따른 영어 마찰음에 대한 한국 어린이 영어 학습자의 인지 능력

임영신 ( Youngshin Lim ),장우혁 ( Wooheok Chang ) 대한언어학회 2019 언어학 Vol.27 No.1

In this study, we investigate how L2 acquisition theories can successfully account for the relative difficulties in the perception of English fricatives by Korean young learners. Regarding the acquisition of L2 phonemes, the previous studies can be divided into two main approaches: (1) phoneme-based analysis which is mainly concerned with Loanword Phonology and Perceptual Assimilation Model vs. (2) feature-based analysis. Noticing that these analyses make different predictions on the acquisition of L2 sounds, we conducted AX discrimination task and 4I-oddity task to find which types of English fricatives (voicing, place, and manner of articulation) Korean young learners have difficulties in perceiving. 30 Korean elementary school students participated in the tasks and 27 minimal pairs including word-initial English fricatives were employed. Crucial to the analysis of the experimental results is that with one particular L2 acquisition theory it is impossible to explain why Korean learners can or cannot perceive English fricatives distinctively. That is, Korean young learners’ perceptibility of English fricatives is attributed to both phonemic differences and feature differences between L1 and L2. Furthermore, in some cases, the fact that Korean young learners fail to discriminate certain pairs of English fricatives should be better accounted for by considering phonetic characteristics of L1 and L2 fricatives.

통신언어의 언어학적 분석

신희삼 ( Hee Sam Shin ) 전북대학교 인문학연구소 2016 건지인문학 Vol.15 No.-

This thesis is a linguistic analysis of the communication language. The communication language is becoming various, complex with time. A linguistic analysis of the communication language is necessary for the specific, prompt communication. For the first time, the use of the communication language is classified into intended and unintended. After that, these are put together for the linguistic analysis. In the linguistic analysis, phonemics, morphemics, syntax, semantics and the others are treated. Orthography, loanword orthography, foreign original language write and so on belong to the others. In addition, the continuity of the communication language is formed four factors. First, the extension of the forms. Second, the emotional homogeneity by substitution. Third, polysemous characteristics. Forth, the intergenerational understanding. Fifth, the suitability of the constituting principle of neologism. Lastly, the communication language has two remaining issues. One is the development of the valid checking model, another is the development of the efficient teaching method.

Pronunciation Errors and Strategies by Korean L2 learners of Mongolian

장우혁 한국몽골학회 2014 몽골학 Vol.0 No.36

The main purpose of this paper is to offer a typological observation ofpossible errors in the pronunciation of Mongolian sounds by Koreanlearners of Mongolian. It is examined that pronunciation errors are mainlyattributed to different phonemic inventories and phonological featuresbetween the Korean language (native language) and the Mongolianlanguage (target language). According to contrastive analysis, Mongolianvoiced and palatalized stops are the most difficult sounds for Koreanlearners to acquire because there is no corresponding segments in Korean. With regard to laryngeal features, Korean obstruents are contrastivedepending on the markedness of Glottal Width (GW) whereas Mongolianobstruents are marked with either Glottal Width (GW) or Glottal Tension(GT). Thus, Mongolian laryngeal system is more complicated than Koreanone. This may induce Korean learners to have difficulty in learningMongolian obstruents. In addition to the segmental differences betweenthe two languages, suprasegmental factors, such as syllable types andvowel length, can prevent Korean learners from acquiring the pronunciationof Mongolian sounds. For example, Mongolian allows a consonant clusterin coda position, which may lead Korean learners to insert epentheicvowels in between adjacent consonants. In Mongolian, vowel length is adistinctive feature to distinguish long vowels from short ones. However,the distinction between long and short vowels is insignificant in Korean. This paper also addresses that the Grammar-Translation method and Audio-lingual method that are popular in teaching foreign languages donot help Korean learners learn how to pronounce Mongolian phonemes. An alternative teaching method, the Silent Way seems to have benefitedmost by associating sounds with certain symbols, which allows Koreanlearners to facilitate an insight into Mongolian phonemes.

24-36개월 아동의 조음 변이성 연구

김희윤(Hee Yun Kim),하승희(Seung hee Ha) 한국언어청각임상학회 2016 Communication Sciences and Disorders Vol.21 No.2

배경 및 목적: 24-36개월 아동의 단어 변이성과 음소 변이성을 자발화 상황에서 측정하여, 아동의 월령 집단 간 차이 비교를 통해 초기 음운 발달과정 중에 나타나는 조음 변이성의 특징을 살펴보고자 하였다. 방법: 24-30개월과 31-36개월 일반아동 각 11명씩 총 22명의 자발화를 수집하여 단어 변이성, 음소 변이성, 자음정확도, 자음목록수를 구하였다. 자발화 수준에서 단어 변이성은 어절 단위로 분석하였고, 정조음과 조음오류를 모두 포함하여 단어 및 음소 변이성을 측정하였다. 두 월령 집단 간에 단어 변이성, 음소 변이성에서 차이가 유의한지 분석하였고, 단어 변이성 및 음소 변이성과 자음정확도 및 자음목록수 간에 상관관계를 살펴보았다. 결과: 30분간 아동이 산출한 발화를 기준으로 단어 변이성을 살펴보면 2세 전반에는 2회 이상 산출한 어절 중 51.55%가 다양한 형태로 산출되다가 2세 후반에는 36.45%로 단어 변이성이 감소하였다. 음소 변이성을 살펴보면 2세 전반에는 아동의 자음목록 중 69%의 음소가 다양한 형태로 산출되다가 2세 후반에는 52.45%로 음소 변이성이 감소하였다. 단어 변이성과 음소 변이성은 모두 자음정확도와 자음목록수 간에 음의 상관관계를 나타냈다. 논의 및 결론: 2세에 음운습득이 비약적으로 이루어지면서 안정적이고 일관된 조음 패턴을 나타낸다는 조음 특성을 제시하고 있다. 본 연구결과는 말소리 발달지연 아동이나 비일관적인 음운 장애를 보이는 아동의 조음 변이성 정도를 보다 객관적으로 비교하고 평가하는 데 사용될 수 있을 것이다. Objectives: This study aims to investigate the articulatory variability of word and phoneme production on spontaneous utterance samples in typically developing children age 24 to 36 months. Methods: Speech samples were collected from 11 children age 24 to 30 months and 11 children age 31 to 36 months. Word variability, phoneme variability, percentage of consonants correct (PCC), and consonant inventory were measured. Eojeol was a basic unit for analysis of word variability. Both correct articulations and error productions were in-cluded for the variability analysis. Statistical analyses were conducted to investigate whether word and phoneme variability measures were different between the two age groups and if there were relationships between articulatory variability measures and PCC and conso-nant inventory. Results: For the children age 24 to 30 months, 51.55% of Eojeols which were produced more than twice showed variable productions and 69% of consonants in the consonant inventory showed variability. For children age 31 to 36 months, word vari-ability and phoneme variability were 36.45% and 52.45%, respectively. Both variability measures showed significantly negative correlation with PCC and consonant inventory. Conclusion: This study suggests that 2-year-old children show progressively stable and consistent articulation patterns along with phonological acquisition. The results of this study have implications for evaluating the degree of articulatory variability in children who have inconsistent phonological disorders or childhood apraxia of speech.

한국어 유의미 단음절의 초·중·종성 주파수 분석 연구

김진숙,신은영,조은빛 한국청각언어재활학회 2013 Audiology and Speech Research Vol.9 No.2

The purpose of this study was to analyse the frequency characteristics of 583 Korean meaningful monosyllabic words by their initial·middle·final phonemes to understand speech recognition ability based on hearing configuration more accurately. Normal hearing 20 Korean young adults(10 males and 10 females) produced 583 words in context situations. Data were collected by using Computerized Speech Lab(CSL 4300B) and analyzed by using Pratt 4.3.14. According to 19 initial consonants, /ㄱ/, /ㄴ/, /ㄷ/, /ㄹ/, /ㅁ/, /ㅂ/, /ㅅ/, /ㅇ/, /ㅈ/, /ㅊ/, /ㅋ/, /ㅌ/, /ㅍ/, /ㅎ/, /ㄲ/, /ㄸ/, /ㅃ/, /ㅆ/, /ㅉ/, 18 middle vowels, /ㅏ/, /ㅐ/, /ㅑ/, /ㅓ/, /ㅔ/, /ㅕ/, /ㅖ/, /ㅗ/, /ㅘ/, /ㅙ/, /ㅚ/, /ㅛ/, /ㅜ/, /ㅝ/, /ㅟ/, /ㅠ/, /ㅡ/, /ㅣ/, 7 and no final consonants, /ㄱ/, /ㄴ/, /ㄷ/, /ㄹ/, /ㅁ/, /ㅂ/, /ㅇ/ were analyzed and their means, standard deviations, and ranges were recorded. The means of initial consonants /ㄴ/, /ㄹ/, /ㅁ/ were recorded under 500 Hz, /ㄱ/, /ㄲ/, /ㅂ/, /ㅃ/, /ㅇ/, /ㅋ/, /ㅍ/, /ㅎ/ were recorded at 1,000 ~ 2,500 Hz, /ㄷ/, /ㄸ/, /ㅈ/, /ㅉ/, /ㅊ/, /ㅌ/ were recorded at 4,000 ~ 5,000 Hz, /ㅅ/, /ㅆ/ were recorded at 5,000 ~ 6,000 Hz. The means of middle vowels /ㅛ/, /ㅝ/, /ㅠ/ were recorded under 1,000 Hz, /ㅑ/, /ㅕ/, /ㅖ/, /ㅗ/, /ㅜ/ were recorded1,000 ~ 2,500 Hz, /ㅏ/, /ㅐ/, /ㅓ/, /ㅔ/, /ㅘ/, /ㅙ/, /ㅚ/, /ㅟ/, /ㅡ/, /ㅣ/ were recorded 2,500 ~ 4,000 Hz, showing no means over 4,000 Hz. The means of 7 and no final consonants /ㄱ/, /ㄴ/, /ㄷ/, /ㄹ/, /ㅁ/, /ㅂ/, /ㅇ/ were recorded at 2,000 ~ 3,000 Hz. The statistical significance using oneway analysis of variance revealed that the initial consonants’ means were different according to their middle vowels, the middle vowels’ means were different according to their initial consonants, and the final consonants’ means were different according to their initial consonants except initial consonants /ㄴ/, /ㄹ/, /ㅁ/. The initial consonants could be categorized by articulation manner and place. These results can be used clinically for better audiological assessment and rehabilitation such as speech sound audiogram of English. In the future, more data should be investigated with female and male talker vocalizations for more systematic and complete frequency analysis according to initial·middle·final phonemes.

현대 그리스어 발음 교육을 위한 기초연구

김혜진 국제언어인문학회 2021 인문언어 Vol.23 No.1

The aim of this study is two-fold. The first is to contrast segmental sounds of Modern Greek and Korean for Korean learners of the Greek language. The second part is some pronunciation instruction suggestions. Our study resulted in an account of specific instances in which pronunciation errors occurred. The first is pronunciations of similar segmental sounds which exist in Korean, but have differences in the way they are articulated. Greek distinguishes plosives (/p/, /b/, /t/, /d/, /k/, /g/) and fricatives (/f/, /v/, / θ/, /ð/) according to the vibration of the vocal cords while Korean distinguishes consonants from aspiration and tension. The second is pronunciation of sounds that do not exist in Korean language (/f/, /v/, /θ/, /ð/, /ɾ/). Third, it is pronunciation of allophones. Korean learners sometimes do not recognize the differences depending on the phonological environment when many allophones exist in a single phoneme. There is a tendency to pronounce all the allophones the same as the representative phoneme (e.g., [ɣ] and [ʝ] as [ ] in Korean equating [ ] and /ɣ/, and pronounce [Σ] as /s/, which should sound differently by /s/ and /z/, depending on the phonological environment). In the case of vowels, the sounds of all five vowels of Greek (/i/, /ɛ/, /ɐ/, /o/, /u/) exist in Korean. Therefore, pronunciation of the segmental sound itself is not expected to be difficult, and errors in vowels are likely to occur in diphthongs and stress. In the part of pronunciation instruction, we suggest implicit pronunciation instruction based on the beginner’s level instructions for the Greek proficiency test, focusing on segmental sounds /θ/, /ð/, /ɣ/ and its allophones [ɣ] and [ʝ]. The elements influencing pronunciation include not only segmental sounds but also suprasegmental elements. However, this study could be a starting point for Modern Greek pronunciation research on acquisition and teaching, given that they partly interact with each other.

The Effects of Phonemic Awareness Instructions on L2 Listening Comprehension: A Meta-Analysis

Seyeon Choe,KyungA Lee,Youngsoon So 아시아영어교육학회 2020 The Journal of Asia TEFL Vol.17 No.4

This study examines the effect of phonemic awareness instruction on listening comprehension ability in learning English as a second language or foreign language (L2). The searching procedures were carried out using Publish or Perish with discreetly selected keywords. Eight studies with 13 samples published between 2000 and 2020 were obtained. Hedges’ g was calculated, leveraging Comprehensive Meta-Analysis (CMA) software. The overall effect size of phonemic awareness instruction on listening skills was found to be large (Hedges’ g = 0.99, lower bound = 0.82, upper bound = 1.61). The result demonstrated that phonemic awareness instructions to beginners (Hedges’ g = 0.86) or primary school students (Hedges’ g = 3.67) might have a large effect on enhancing their listening skills. Furthermore, research that conducted phonemic awareness instructions (Hedges’ g = 1.43) showed a much larger effect size than phonics instructions (Hedges’ g = 0.45), suggesting the need for a more focused phonemic instruction for L2 learners. Lastly, the effectiveness of phonemic awareness instruction was better assessed when using intensive listening measurements (Hedges’ g = 1.43) rather than selective ones (Hedges’ g = 0.80). These results collectively indicate that phonemic awareness instructions could be both effective and practical for helping L2 learners improve listening skills.

한국어 음소분석에 관한 연구

이수길, 송정영 배재대학교 공학연구소 2014 공학논문집 Vol.16 No.1

It is generally known that accurate segmentation is very necessary for both an individual word and continuous utterances in speech recognition. It is also commonly known that techniques are now being developed to classify the voiced and the unvoiced, also classifying the plosives and the fricatives. The method for accurate recognition of the phonemes isn't yet scientifically established. Therefore, in this study we analyze the Korean language, using the classification of "Hunminjeongeum" and contemporary phonetics, with the frequency band, Mel band and Mel Cepstrum, we extract notable features of the phonemes from Korean speech and segment speech by the unit of the phonemes to normalize them. Finally, through the analysis and verification, we intend to set up Phonemic Segmentation System that will make us able to adapt it to both an individual word and continuous utterances.

영어 음소 인식 지도에 대한 메타 분석

김세진(Kim, Se-jin),김정렬(Kim, Jeong-ryeol) 한국영어교과교육학회 2015 영어교과교육 Vol.14 No.3

A quantitative meta-analysis evaluating the effects of English phonemic awareness (PA) instruction on reading, writing, and affective domain outcomes for Korean EFL learners was conducted. A meta-analysis of 34 PA studies showed that the overall effect size of PA instruction was significant (.836). PA instruction exerted a statistically significant impact on phonemic awareness(1.073), reading(.675), writing(.817) and affective domain outcome(.752) for Korean primary and kindergarten learners. Moderator variables analysis also showed significant results. First, PA instruction for 6-9 year old learners (.929) was more effective than for 10-11 or 12-13 year olds. Second, PA activities using nursery rhymes, songs, poems, and chants characterized by rhythm and melody (.869) had more positive effects than games or storytelling do. Third, two PA tasks (.973) were more effective than one or more than three tasks. Fourth, rhyme and onset PA tasks (1.047) were more effective than just rhyming tasks or various PA tasks. Lastly, PA instruction lasting more than 10 hours (.946) was more effective than 1 to 4.5 hour or 5 to 9.5 hours periods of instruction. In sum, PA instruction had a significant effect on developing Korean learners’ English phonemic awareness, literacy (reading and writing abilities), and affective domain for learning English.