• Title/Summary/Keyword: Comparable corpora

Search Result 6, Processing Time 0.018 seconds

Bilingual lexicon induction through a pivot language

  • Kim, Jae-Hoon;Seo, Hyeong-Won;Kwon, Hong-Seok
    • Journal of Advanced Marine Engineering and Technology
    • /
    • v.37 no.3
    • /
    • pp.300-306
    • /
    • 2013
  • This paper presents a new method for constructing bilingual lexicons through a pivot language. The proposed method is adapted from the context-based approach, called the standard approach, which is well-known for building bilingual lexicons using comparable corpora. The main difference between the standard approach and the proposed method is how to represent context vectors. The former is to represent context vectors in a target language, while the latter in a pivot language. The proposed method is very simplified from the standard approach thereby. Furthermore, the proposed method is more accurate than the standard approach because it uses parallel corpora instead of comparable corpora. The experiments are conducted on a language pair, Korean and Spanish. Our experimental results have shown that the proposed method is quite attractive where a parallel corpus directly between source and target languages are unavailable, but both source-pivot and pivot-target parallel corpora are available.

A Corpus-based Study of Translation Universals in English Translations of Korean Newspaper Texts (한국 신문의 영어 번역에 나타난 번역 보편소의 코퍼스 기반 분석)

  • Goh, Gwang-Yoon;Lee, Younghee (Cheri)
    • Cross-Cultural Studies
    • /
    • v.45
    • /
    • pp.109-143
    • /
    • 2016
  • This article examines distinctive linguistic shifts of translational English in an effort to verify the validity of the translation universals hypotheses, including simplification, explicitation, normalization and leveling-out, which have been most heavily explored to date. A large-scale study involving comparable corpora of translated and non-translated English newspaper texts has been carried out to typify particular linguistic attributes inherent in translated texts. The main findings are as follows. First, by employing the parameters of STTR, top-to-bottom frequency words, and mean values of sentence lengths, the translational instances of simplification have been detected across the translated English newspaper corpora. In contrast, the portion of function words produced contrary results, which in turn suggests that this feature might not constitute an effective test of the hypothesis. Second, it was found that the use of connectives was more salient in original English newspaper texts than translated English texts, being incompatible with the explicitation hypothesis. Third, as an indicator of translational normalization, lexical bundles were found to be more pervasive in translated texts than in non-translated texts, which is expected from and therefore support the normalization hypothesis. Finally, the standard deviations of both STTR and mean sentence lengths turned out to be higher in translated texts, indicating that the translated English newspaper texts were less leveled out within the same corpus group, which is opposed to what the leveling-out hypothesis postulates. Overall, the results suggest that not all four hypotheses may qualify for the label translation universals, or at least that some translational predictors are not feasible enough to evaluate the effectiveness of the translation universals hypotheses.

A Protein-Protein Interaction Extraction Approach Based on Large Pre-trained Language Model and Adversarial Training

  • Tang, Zhan;Guo, Xuchao;Bai, Zhao;Diao, Lei;Lu, Shuhan;Li, Lin
    • KSII Transactions on Internet and Information Systems (TIIS)
    • /
    • v.16 no.3
    • /
    • pp.771-791
    • /
    • 2022
  • Protein-protein interaction (PPI) extraction from original text is important for revealing the molecular mechanism of biological processes. With the rapid growth of biomedical literature, manually extracting PPI has become more time-consuming and laborious. Therefore, the automatic PPI extraction from the raw literature through natural language processing technology has attracted the attention of the majority of researchers. We propose a PPI extraction model based on the large pre-trained language model and adversarial training. It enhances the learning of semantic and syntactic features using BioBERT pre-trained weights, which are built on large-scale domain corpora, and adversarial perturbations are applied to the embedding layer to improve the robustness of the model. Experimental results showed that the proposed model achieved the highest F1 scores (83.93% and 90.31%) on two corpora with large sample sizes, namely, AIMed and BioInfer, respectively, compared with the previous method. It also achieved comparable performance on three corpora with small sample sizes, namely, HPRD50, IEPA, and LLL.

Bilingual Lexicon Extraction Using Self-Organizing Maps (자기조직화 지도를 이용한 이중언어사전 자동 구축)

  • Seo, Hyeong-Won;Cheon, Minah;Kim, Jae-Hoon
    • Proceedings of the Korea Information Processing Society Conference
    • /
    • 2015.04a
    • /
    • pp.802-805
    • /
    • 2015
  • 본 논문은 인공신경망(artificial neural network)의 한 종류인 자기조직화 지도(self-organizing map)를 이용하여 비교말뭉치(comparable corpora)로부터 이중언어사전(bilingual lexicon)을 자동으로 구축하는 방법에 대하여 기술한다. 일반적으로 우리가 대상으로 하는 언어 쌍마다 말뭉치 혹은 초기사전과 같은 언어 자원을 수집하고 그것을 필요에 맞게 가공하는 것은 매우 어려운 일이다. 이런 관점에서 볼 때, 비지도학습(unsupervised learning) 방법 중 하나인 자기조직화 지도를 이용하여 사전을 구축하면 다른 방법에 비해 적은 노력으로도 더 높은 성능을 얻을 수 있다. 본 논문에서는 한국어와 불어에 대하여 실험을 하였고, 그 결과 적은 양의 초기사전으로도 주목할 만한 정확도를 얻을 수 있었다. 향후 연구로는 학습 파라미터에 대해 좀 더 다양한 실험을 하고, 다른 언어 쌍으로의 적용 및 기존의 평가사전을 확장하여 더 많은 경우에 대해 실험하는 것을 들 수 있다.

Effect of Genistein on the Onset of Puberty in Female Rats (암컷 흰쥐의 사춘기 개시에 미치는 Genistein의 효과)

  • Lee, Kyeung-Yeup;Lee, Sung-Ho
    • Development and Reproduction
    • /
    • v.10 no.1
    • /
    • pp.55-61
    • /
    • 2006
  • There is growing concern that dietary soy intake is associated with protection of breast cancer. However, questions persist on the potential adverse effects of the main soy constituent genistein(GS) on female reproductive physiology. In this study, we examined whether prepubertal exposure to GS affected on the onset of puberty and the associated reproductive parameters such as hormone receptor expressions in female rats. GS(100mg/kg/day) was administrated daily from postnatal day 25(PND 25) to the day when the first vaginal opening(VO) was observed, and the animals were sacrificed on the day after VO occurred. Gross anatomy and tissue weight were compared to test the GS's effect on the cell proliferation. Furthermore, histological studies were performed to assess the structural alterations in tissues. Specific radioimmunoassay(RIA) were carried out to measure serum LH levels. To determine the transcriptional changes in progesterone receptors(PR), total RNAs were extracted from ovary and uterus and were applied to semi-quantitative reverse transcription polymerase chain reaction(RT-PCR). As a results, advanced VO was shown in the GS group(PND $31.2{\pm}0.6$) compared to the vehicle group (PND $35.3{\pm}0.7$). GS treatment significantly increased wet weight of ovaries and uteri compared to the vehicle group. Increased serum LH levels were also shown in the GS group. Graafian follicles and corpora lutea(CL) were observed only in the ovaries from GS treated animals. Similarly, hypertrophy of luminal and glandular uterine epithelium were found only in the GS group. Collectively, these effects were probably due to the estrogenic effects of GS. In the semi-quantitative RT-PCR studies, the transcriptional activities of PR in both ovary and uterus from GS-treated group were significantly higher than those from the vehicle group. The present studies demonstrated that acute exposure to GS, at levels comparable to the ranges of human exposure, during the critical period of prepubertal stage activates the reproductive system resulting precocious puberty in immature female rats.

  • PDF

KB-BERT: Training and Application of Korean Pre-trained Language Model in Financial Domain (KB-BERT: 금융 특화 한국어 사전학습 언어모델과 그 응용)

  • Kim, Donggyu;Lee, Dongwook;Park, Jangwon;Oh, Sungwoo;Kwon, Sungjun;Lee, Inyong;Choi, Dongwon
    • Journal of Intelligence and Information Systems
    • /
    • v.28 no.2
    • /
    • pp.191-206
    • /
    • 2022
  • Recently, it is a de-facto approach to utilize a pre-trained language model(PLM) to achieve the state-of-the-art performance for various natural language tasks(called downstream tasks) such as sentiment analysis and question answering. However, similar to any other machine learning method, PLM tends to depend on the data distribution seen during the training phase and shows worse performance on the unseen (Out-of-Distribution) domain. Due to the aforementioned reason, there have been many efforts to develop domain-specified PLM for various fields such as medical and legal industries. In this paper, we discuss the training of a finance domain-specified PLM for the Korean language and its applications. Our finance domain-specified PLM, KB-BERT, is trained on a carefully curated financial corpus that includes domain-specific documents such as financial reports. We provide extensive performance evaluation results on three natural language tasks, topic classification, sentiment analysis, and question answering. Compared to the state-of-the-art Korean PLM models such as KoELECTRA and KLUE-RoBERTa, KB-BERT shows comparable performance on general datasets based on common corpora like Wikipedia and news articles. Moreover, KB-BERT outperforms compared models on finance domain datasets that require finance-specific knowledge to solve given problems.