Ambiguous Sentence Processing in Translation

Activation of lexical and syntactic target language properties in translation

Lexical and syntactic target language interactions in translation

Previous works on cross-lingual Named Entity Recognition (NER) have achieved great success. However, few of them consider the effect of language families between the source and target languages. In this study, we find that the cross-lingual NER performance of a target language would decrease when its source language is changed from the same (homogenous) into a different (heterogenous) language family with that target language. To improve the NER performance in this situation, we propose a novel cross-lingual NER framework based on self-distillation mechanism and Bilateral-Branch Network (SD-BBN). SD-BBN learns source-language NER knowledge from supervised datasets and obtains target-language knowledge from weakly supervised datasets. These two kinds of knowledge are then fused based on self-distillation mechanism for better identifying entities in the target language. We evaluate SD-BBN on 9 language datasets from 4 different language families. Results show that SD-BBN tends to outperform baseline methods. Remarkably, when the target and source languages are heterogenous, SD-BBN can achieve a greater boost. Our results might suggest that obtaining language-specific knowledge from the target language is essential for improving cross-lingual NER when the source and target languages are heterogenous. This finding could provide a novel insight into further research.

In this study, transfer learning techniques are presented for cross-lingual speech recognition to mitigate the effects of limited availability of data in a target language using data from richly resourced source languages. First, a maximum likelihood (ML) based regularization criterion is used to learn context-dependent Gaussian mixture model (GMM) based hidden Markov model (HMM) parameters for phones in target language using data from both target and source languages. Recognition results indicate improved HMM state alignments. Second, the hidden layers of a deep neural network (DNN) are initialized using unsupervised pre-training of a multilingual deep belief network (DBN). The DNN is fine-tuned jointly using a modified cross entropy criterion that uses HMM state alignments from both target and source languages. Third, another DNN fine-tuning technique is explored where the training is performed in a sequential manner source language followed by the target language. Experiments conducted using varying amounts of target data indicate further improvements in performance can be obtained using joint and sequential training of the DNN compared to existing techniques. Turkish and English were chosen to be the target and source languages respectively.

This paper discusses the metaphor categories and the translation strategies that are used in Twilight (2005) novel using Newark’s (1988) theory of metaphor categories and Larson’s (1984) theory of metaphor translation strategies. The acquired data consist of verbal data from Source Language (SL) text and Target Language (TL) text from both novels. This study is a qualitative study with note-taking technique library research, and observation. The library research and observation were applied to obtain information and evidence about the metaphor categories and the translation strategies used by the translator. The data then collected and noted using the note-taking technique, and there were 264 data gathered from SL and TL text which then analyzed to determine the categories and translation strategies of the metaphors. The analysis of the data then showed that dead metaphor is the main category that are found in SL and TL text with 172 data in SL text and 136 data in TL text. The next categories found is stock metaphor with 61 data in SL and 36 data in TL text, followed by cliché metaphor with 31 data in SL and 26 data in TL. There were no original, adapted, and recent metaphor categories found from the gathered data. In terms of metaphor translation strategies, there are four strategies used by the translator to translate metaphor as follows: (1) 150 data found translated using the strategy of metaphor/simile to metaphor/simile with the same image, (2) 9 data found translated using metaphor to simile strategy, (3) 56 data found translated by using metaphor to metaphor in the receptor language strategy, and (4) 150 data found translated using the strategy of metaphor to non-figurative language. The determination of metaphor categories and the translation strategies can help both the translator and the receptor language readers in a way that the categorization of the metaphor may assist the translator consider the most appropriate strategy to translate figurative language. Moreover, using the appropriate translation strategy can contribute to the accuracy, acceptability, and readability of the translation receptor language readers.

(ProQuest: ... denotes non-US-ASCII text omitted.)1 IntroductionThe aim of the paper is study different ways of using specific lexica and different methods of its translation. The paper examines which extent and under which conditions a translator is able to liberate the language imprisoned in a in his re-creation of that work (Benjamin 2008:82). As Brisset (2008:337) argues, is a dual act of communication. It presupposes the existence, not of a single code, but of two distinct codes, the'source language' and the 'target language'. The fact that the two codes are not isomorphic creates obstacles for the translative operation. This explains why linguistic questions are the starting-point for all thinking about translation.In some cases, the intrinsic quality of the text (Xie 2006:207) cannot be preserved without transposition of linguistic and cultural patterns or culture units. The paper gives the examples of translating (or preserving) specific phrases and lexical items as the superficial markers of overlapping of different cultures, ethnicities, beliefs and values.In the analysis of the translation procedures we will apply definitions of cultural information given by several authors. Chick (1999:5) comments that When viewed as information, includes shared knowledge, behavioural patterns, material artefacts and may be characteristic of either small groups of individuals or of large aggregates of people. The first of the prerequisites which Durham (1991:188) sets in order define units of is that the cultural unit of information that actually or potentially guides behaviour. Galisson (1991:116) makes the difference between two types of culture: explicit, learned which consists of encyclopedic, learnable knowledge and implicit culture, common culture. The 'lexiculture' is defined as the entirety of culture-bound and frequent lexical items, which are more or less culture-loaded.2 In other words, 'lexiculture' is a collection of implicit knowledge, shared by all members of a linguistic community.The signs of the lexiculture, or more widely, the units of culture, can be used for different aims in the text. Translation strategies should respond these aims. As is a special kind of transfer, one of which is culturally determined, since forms a general framework within which any possible action is included (Inigo Ros 2003:4), the paper aims analyse translation strategies of source texts which contain different cultural references themselves.It is well known that the translator must have perfect knowledge both of the source and the target cultures: 'Er muss bikulturell sein' (Reiss & Vermeer 1984:26) (Inigo Ros 2003:4). But, when the source text already contains a variety of cultural references from culture(s) other than the source culture, it is not easy either for the translator or for the reader of the target text follow the multicultural references and mediate the complicated and multiplied cultural transfers - primarily from the foreign culture(s) the source text, and secondarily from the source the target text.In such cases, the role of foreign lexica and its transference the target text is very important. Moreover, one of the groups of such lexical items is that of 'cultural words'. Despite the fact that there is no absolute agreement on defining the concept of 'cultural words',3 there is a general agreement that such words denote (at least) material objects which are characteristic for certain nations (i.e. for their customs, culture, social and historical development) (Vlahov & Florin 1980:47-140). Cultural words can also be connected different phases of the social and historical development of a community (Kutz 1977:256). In other words, these lexical items belong certain semantic fields.4The treatment of untranslatable culture-specific words is one of the important questions in this paper. …

This study focused on the analysis of the translation strategies and resulted translation quality in rendering metaphors found in the Twilight novel. The analysis involved the translation strategies by Newmark (1988) and translation quality by Hartono (2016). The results showed that there were seven translation strategies utilized by the translator in rendering the metaphors in the Twilight novel from Indonesian into English with total metaphors were 164. Those metaphors were translated using translation strategies: reproducing the same metaphorical image in the target language 48 times (29%), replacing the metaphorical image in the source language with a standard metaphorical image in the target language with 25 times (15%), translating metaphor by simile by maintaining the metaphorical image with 20 times (12%), translating metaphor by simile plus sense with 13 times (8%), converting a metaphor to its sense or meaning only with 44 times (29%), deleting metaphor with 7 times (4%), and translating metaphor by the same metaphor with the sense or meaning added with 7 times (4%). In terms of content quality, the translation of metaphors was dominated by good content, presentation, and mechanics quality. Moreover, the application of each translation strategy would result in different types of translation quality in terms of content, presentation and mechanics. In addition, it is suggested that the employment of translation strategy in rendering metaphor should consider deeply the existence of the same metaphor both in the source language and target language because the quality of the translation will be determined by the translation strategies.

There is no target language can fully represent the source language. This study focuses on the problem of differences, uniqueness and limitations of Sundanese as the target language in accommodating Arabic Quran, using the linguistic approach to translation. I use eleven translations of the Quranic Sundanese asthe mainobject. This study shows that there is a similar structure of Sundanese language in the midst of variations of the translation of the Quran. I confirm that the translation of the Quran in Sundanese has limitations and barriers,such as thechange of lexical, word types and unit of semantic and structure of sentences. The most complicated is translation in the form of Sundanese metrical verse or dangding. The translationsof the Quran in Sundanese, unlike the Bible translationsin Christianity, tend to be fettered by the structure of source language.The Sundanese translators generally difficultto use the structure of target language, because it is bounded by their loyaltiesto the source language. Therefore, there are various compromises were taken by translators to bridge the complexity between both target and source language structures. It is an attempt to show inability of Sundaneselanguage to accommodate Arabiclanguage in the midst of rich and complexity of the language

The present work investigates the relationship between syntactic variation and priming in translation. It is based on the claim that languages share a common cognitive network of neural activity. When the source and target languages are solicited in a translation context, this shared network can lead to facilitation effects, so-called priming effects. We suggest that priming is a default setting in translation, a special case of language use where source and target languages are constantly co-activated. Such priming effects are not restricted to lexical elements, but do also occur on the syntactic level. We tested these hypotheses with translation data from the TPR database, more specifically for three language pairs (English-German, English-Danish, and English-Spanish). Our results show that response times are shorter when syntactic structures are shared. The model explains this through strongly co-activated network activity, which triggers a priming effect.

This study aims to find out in-depth about the application of foreignization strategies in translating the Quran into Indonesian. The application of foreignization strategies in terms of meaning accuracy in the source language (SL) and target language (TL) as well as its suitability with the cultural context. The research method used is content analysis adapted from Krippendorff (1980). Based on an analysis of the foreignization strategy implemented by the Quran Translation Team of the Ministry of Religious Affairs (MoRA) of the Republic of Indonesia, it was found that 94 words and phrases were foreignized in the translation of Surah Al-Baqarah. Judging from the aspect of the accuracy of meaning, equivalence, and its use in Indonesian, the vocabulary is still accurate in meaning, as intended by the SL text and its users. However, some absorption vocabularies experience a shift in meaning, broad, narrow, general, and specific, as well as those containing good and bad values. Two factors influence the translator in implementing the foreignization strategy in the translation of the Quran, namely (a) the condition of the Quranic text, which is difficult to translate into Indonesian, and (b) the translator’s belief that foreign vocabulary retention (foreignization) can give the nuances of the authenticity of Arabic culture (Quran).

Subtitle translation is a little bit different from general translation. Technically, subtitle translation has a lot of limitations, for example, space limits and time constraints. In addition, the differences between the source language and the target language become other obstacles. To solve these obstacles, the translator could use the help of translation strategies. The purpose of this research is to analyze the translation problems existing in a Chinese movie and analyze the result of translation strategies used in its subtitle. To analyze the subtitle in “Ne Zha” (2019), the writer used Gottlieb’s ten translation strategies with the qualitative method. The results show that 28 lines in the subtitles do not match the dialogues and the contexts of the movie. From those 28 lines, paraphrasing is the most frequently used. The reason why paraphrasing is the highest is that there are different grammatical structures between the source language and the target language. Therefore, the translator needs to adjust the Chinese subtitles to Indonesian grammar during translation. However, in the process of the adjustment, the translator not only needs to be able to translate the meaning of the source language into the target language but also needs to understand the plot in the movie to be able to produce good translation results.

Though most translation systems have some mechanism for translating certain types of divergent predicate-argument structures, they do not provide a general procedure that takes advantage of the relationship between lexical-semantic structure and syntactic structure. A divergent predicate-argument structure is one in which the predicate (e.g., the main verb) or its arguments (e.g., the subject and object) do not have the same syntactic ordering properties for both the source and target language. To account for such ordering differences, a machine translator must consider language-specific syntactic idiosyncrasies that distinguish a target language from a source language, while making use of lexical-semantic uniformities that tie the two languages together. This paper describes the mechanisms used by the UNITRAN machine translation system for mapping an underlying lexical-conceptual structure to a syntactic structure (and vice versa), and it shows how these mechanisms coupled with a set of general linking routines solve the problem of thematic divergence in machine translation.

Cross Language Text Categorization (CLTC) is the task of assigning class labels to documents written in a target language (e.g. Chinese) while the system is trained using labeled examples in a source language (e.g. English). With the technique of CLTC, we can build classifiers for multiple languages employing the existing training data in only one language, therefore avoid the cost of preparing training data for each individual language. One challenge for CLTC is the culture differences between languages, which causes the classifier trained on the source language doesn't perform well on the target language. In this paper, we propose an active learning algorithm for CLTC, which takes full advantage of both labeled data in the source language and unlabeled data in the target language. The classifier first learns the classification knowledge from the source language, and then learns the cultural dependent knowledge from the target language. In addition, we extend our algorithm to double viewed form by considering the source and target language as two views of the classification problem. Experiments show that our algorithm can effectively improve the cross language classification performance.

This study investigates the translation problems, strategies, and the most frequently used strategy in translating Indonesian thesis abstracts into English. The subjects of this research comprised eight thesis abstracts written by non-English department students at the University of Mataram during the 2023 academic year. Utilizing Baker’s theory on translation problems and strategies, which encompasses eleven aspects, this study employed a mixed-methods approach, combining quantitative and qualitative analyses. The findings reveal a total of 78 translation problems across the eight Indonesian thesis abstracts. Among Baker's identified problems, only eight were observed, including: culture-specific concepts (31 instances), source language concepts not lexicalized in the target language (4 instances), semantically complex source language words (14 instances), a lack of specific terms in the target language (1 instance), differences in expressive meaning (8 instances), differences in form (2 instances), differences in frequency and purpose of specific forms (6 instances), and the use of loan words in the source text (12 instances). To address these challenges, seven out of eight translation strategies from Baker’s theory were employed in the English thesis abstracts. The most utilized strategy was translation using loan words (35 instances, 45%), followed by translation using a more general word (12 instances, 16%), translation using a neutral word (10 instances, 13%), translation by omission (10 instances, 13%), translation by paraphrase using a related word (4 instances, 5%), and the least employed strategies were translation using cultural substitution (3 instances, 4%) and paraphrase using unrelated words (3 instances, 4%). Of the 78 translation problems identified, 77 were resolved using the seven strategies, while one problem remained unsolved due to an unacceptable translation. This thesis aims to serve as a valuable reference for translators, English department students, and researchers in the field of translation. The findings indicate a significant need for enhanced support and resources in the translation process for non-English department students at the University of Mataram.

Social media serves as a platform for netizens to stay informed and express their opinions through the Internet. Currently, the social media discourse environment faces a significant security threat—offensive comments. A group of users posts comments that are provocative, discriminatory, and objectionable, intending to disrupt online discussions, provoke others, and incite intergroup conflict. These comments undermine citizens’ legitimate rights, disrupt social order, and may even lead to real-world violent incidents. However, current automatic detection of offensive language primarily focuses on a few high-resource languages, leaving low-resource languages, such as Malay, with insufficient annotated corpora for effective detection. To address this, we propose a zero-shot, cross-language unsupervised offensive language detection (OLD) method using a dual-branch mBERT transfer approach. Firstly, using the multi-language BERT (mBERT) model as the foundational language model, the first network branch automatically extracts features from both source and target domain data. Subsequently, Sinkhorn distance is employed to measure the discrepancy between the source and target language feature representations. By estimating the Sinkhorn distance between the labeled source language (e.g., English) and the unlabeled target language (e.g., Malay) feature representations, the method minimizes the Sinkhorn distance adversarially to provide more stable gradients, thereby extracting effective domain-shared features. Finally, offensive pivot words from the source and target language training sets are identified. These pivot words are then removed from the training data in a second network branch, which employs the same architecture. This process constructs an auxiliary OLD task. By concealing offensive pivot words in the training data, the model reduces overfitting and enhances robustness to the target language. In the end-to-end framework training, the combination of cross-lingual shared features and independent features culminates in unsupervised detection of offensive speech in the target language. The experimental results demonstrate that employing cross-language model transfer learning can achieve unsupervised detection of offensive content in low-resource languages. The number of labeled samples in the source language is positively correlated with transfer performance, and a greater similarity between the source and target languages leads to better transfer effects. The proposed method achieves the best performance in OLD on the Malay dataset, achieving an F1 score of 80.7%. It accurately identifies features of offensive speech, such as sarcasm, mockery, and implicit expressions, and showcases strong generalization and excellent stability across different target languages.