Abstract: The low availability of well-trained, unlimited, dynamic-access models for specific languages makes it hard for corporate users to adopt quick translation techniques and incorporate them into product solutions. As translation tasks increasingly require a dynamic sequence learning curve; stable, cost-free opensource models are scarce. We survey and compare current translation techniques and propose a modified sequence to sequence model repurposed with attention techniques. Sequence learning using an encoder-decoder model is now paving the path for higher precision levels in translation. Using a Convolutional Neural Network (CNN) encoder and a Recurrent Neural Network (RNN) decoder background, we use Fairseq tools to produce an end-to-end bilingually trained Spanish-English machine translation model including source language detection. We acquire competitive results using a duo-lingo-corpus trained model to provide for prospective, ready-made plug-in use for compound sentences and document translations. Our model serves a decent system for large, organizational data translation needs. While acknowledging its shortcomings and future scope, it also identifies itself as a well-optimized deep neural network model and solution.
Abstract: This article presents a comparative study evaluating and comparing the quality of machine translation (MT) output of Chinese gastronomy nomenclature. Chinese gastronomic culture is experiencing an increased international acknowledgment nowadays. The nomenclature of Chinese gastronomy not only reflects a specific aspect of culture, but it is related to other areas of society such as philosophy, traditional medicine, etc. Chinese dish names are composed of several types of cultural references, such as ingredients, colors, flavors, culinary techniques, cooking utensils, toponyms, anthroponyms, metaphors, historical tales, among others. These cultural references act as one of the biggest difficulties in translation, in which the use of translation techniques is usually required. Regarding the lack of Chinese food-related translation studies, especially in Chinese-Spanish translation, and the current massive use of MT, the quality of the MT output of Chinese dish names is questioned. Fifty Chinese dish names with different types of cultural components were selected in order to complete this study. First, all of these dish names were translated by three different MT tools (Google Translate, Baidu Translate and Bing Translator). Second, a questionnaire was designed and completed by 12 Chinese online users (Chinese graduates of a Hispanic Philology major) in order to find out user preferences regarding the collected MT output. Finally, human translation techniques were observed and analyzed to identify what translation techniques would be observed more often in the preferred MT proposals. The result reveals that the MT output of the Chinese gastronomy nomenclature is not of high quality. It would be recommended not to trust the MT in occasions like restaurant menus, TV culinary shows, etc. However, the MT output could be used as an aid for tourists to have a general idea of a dish (the main ingredients, for example). Literal translation turned out to be the most observed technique, followed by borrowing, generalization and adaptation, while amplification, particularization and transposition were infrequently observed. Possibly because that the MT engines at present are limited to relate equivalent terms and offer literal translations without taking into account the whole context meaning of the dish name, which is essential to the application of those less observed techniques. This could give insight into the post-editing of the Chinese dish name translation. By observing and analyzing translation techniques in the proposals of the machine translators, the post-editors could better decide which techniques to apply in each case so as to correct mistakes and improve the quality of the translation.
Abstract: Sentiment analysis and opinion mining have become
emerging topics of research in recent years but most of the work
is focused on data in the English language. A comprehensive
research and analysis are essential which considers multiple
languages, machine translation techniques, and different classifiers.
This paper presents, a comparative analysis of different approaches
for multilingual sentiment analysis. These approaches are divided
into two parts: one using classification of text without language
translation and second using the translation of testing data to a
target language, such as English, before classification. The presented
research and results are useful for understanding whether machine
translation should be used for multilingual sentiment analysis or
building language specific sentiment classification systems is a better
approach. The effects of language translation techniques, features,
and accuracy of various classifiers for multilingual sentiment analysis
is also discussed in this study.
Abstract: Syntactic parsing is vital for semantic treatment by many applications related to natural language processing (NLP), because form and content coincide in many cases. However, it has not yet reached the levels of reliable performance. By manually examining and analyzing individual machine translation output errors that involve syntax as well as semantics, this study attempts to discover what is required for improving syntactic and semantic parsing.
Abstract: This research aims at finding out the causes that led to wrong lexical selections in machine translation (MT) rather than categorizing lexical errors, which has been a main practice in error analysis. By manually examining and analyzing lexical errors outputted by a MT system, it suggests what knowledge would help the system reduce lexical errors.
Abstract: Our program compares French and Italian translations of Homer’s Odyssey, from the XVIth to the XXth century. We focus on the third point, showing how distributional semantics systems can be used both to improve alignment between different French translations as well as between the Greek text and a French translation. Although we focus on French examples, the techniques we display are completely language independent.
Abstract: The present study is an attempt to provide a relatively
comprehensive preview of the Iranian English translators’ perception
on Machine Translation. Furthermore, the study tries to shed light on
the status of implementation of Machine Translation among the
Iranian English Translators. To reach the aforementioned objectives,
the Localization Industry Standards Association’s questioner for
measuring perceptions with regard to the adoption of a technology
innovation was adapted and used to investigate the perception and
implementation of Machine Translation applications by the Iranian
English language translators. The participants of the study were 224
last-year undergraduate Iranian students of English translation at 10
universities across the country. The study revealed a very low level of
adoption and a very high level of willingness to get familiar with and
learn about Machine Translation, as well as a positive perception of
and attitude toward Machine Translation by the Iranian English
translators.
Abstract: The paper presents combined automatic speech
recognition (ASR) of English and machine translation (MT) for
English and Croatian and Croatian-English language pairs in the
domain of business correspondence. The first part presents results of
training the ASR commercial system on English data sets, enriched
by error analysis. The second part presents results of machine
translation performed by free online tool for English and Croatian
and Croatian-English language pairs. Human evaluation in terms of
usability is conducted and internal consistency calculated by
Cronbach's alpha coefficient, enriched by error analysis. Automatic
evaluation is performed by WER (Word Error Rate) and PER
(Position-independent word Error Rate) metrics, followed by
investigation of Pearson’s correlation with human evaluation.
Abstract: Machine Translation (MT) between the Thai and English languages has been a challenging research topic in natural language processing. Most research has been done on English to Thai machine translation, but not the other way around. This paper presents a Thai to English Machine Translation System that translates a Thai sentence into interlingua of a Thai LFG tree using LFG grammar and a bottom up parser. The Thai LFG tree is then transformed into the corresponding English LFG tree by pattern matching and node transformation. Finally, an equivalent English sentence is created using structural information prescribed by the English LFG tree. Based on results of experiments designed to evaluate the performance of the proposed system, it can be stated that the system has been proven to be effective in providing a useful translation from Thai to English.
Abstract: Word sense disambiguation is one of the most important open problems in natural language processing applications such as information retrieval and machine translation. Many approach strategies can be employed to resolve word ambiguity with a reasonable degree of accuracy. These strategies are: knowledgebased, corpus-based, and hybrid-based. This paper pays attention to the corpus-based strategy that employs an unsupervised learning method for disambiguation. We report our investigation of Latent Semantic Indexing (LSI), an information retrieval technique and unsupervised learning, to the task of Thai noun and verbal word sense disambiguation. The Latent Semantic Indexing has been shown to be efficient and effective for Information Retrieval. For the purposes of this research, we report experiments on two Thai polysemous words, namely /hua4/ and /kep1/ that are used as a representative of Thai nouns and verbs respectively. The results of these experiments demonstrate the effectiveness and indicate the potential of applying vector-based distributional information measures to semantic disambiguation.
Abstract: Selecting the word translation from a set of target
language words, one that conveys the correct sense of source word
and makes more fluent target language output, is one of core
problems in machine translation. In this paper we compare the 3
methods of estimating word translation probabilities for selecting the
translation word in Thai – English Machine Translation. The 3
methods are (1) Method based on frequency of word translation, (2)
Method based on collocation of word translation, and (3) Method
based on Expectation Maximization (EM) algorithm. For evaluation
we used Thai – English parallel sentences generated by NECTEC.
The method based on EM algorithm is the best method in comparison
to the other methods and gives the satisfying results.
Abstract: Machine Translation, (hereafter in this document
referred to as the "MT") faces a lot of complex problems from its
origination. Extracting multiword expressions is also one of the
complex problems in MT. Finding multiword expressions during
translating a sentence from English into Urdu, through existing
solutions, takes a lot of time and occupies system resources. We have
designed a simple relational data approach, in which we simply set a
bit in dictionary (database) for multiword, to find and handle
multiword expression. This approach handles multiword efficiently.
Abstract: In this paper, we propose a new model of English-
Vietnamese bilingual Information Retrieval system. Although there
are so many CLIR systems had been researched and built, the accuracy of searching results in different languages that the CLIR
system supports still need to improve, especially in finding bilingual documents. The problems identified in this paper are the limitation of
machine translation-s result and the extra large collections of document to be found. So we try to establish a different model to overcome these problems.
Abstract: It is an important task in Korean-English machine
translation to classify the gender of names correctly. When a sentence
is composed of two or more clauses and only one subject is given as a proper noun, it is important to find the gender of the proper noun
for correct translation of the sentence. This is because a singular pronoun has a gender in English while it does not in Korean. Thus,
in Korean-English machine translation, the gender of a proper noun should be determined. More generally, this task can be expanded into the classification of the general Korean names. This paper proposes a statistical method for this problem. By considering a name as just
a sequence of syllables, it is possible to get a statistics for each name from a collection of names. An evaluation of the proposed method
yields the improvement in accuracy over the simple looking-up of the
collection. While the accuracy of the looking-up method is 64.11%, that of the proposed method is 81.49%. This implies that the proposed
method is more plausible for the gender classification of the Korean names.
Abstract: Many measures have been proposed for machine
translation evaluation (MTE) while little research has been done on
the performance of MTE methods. This paper is an effort for MTE
performance analysis. A general frame is proposed for the description
of the MTE measure and the test suite, including whether the
automatic measure is consistent with human evaluation, whether
different results from various measures or test suites are consistent,
whether the content of the test suite is suitable for performance
evaluation, the degree of difficulty of the test suite and its influence
on the MTE, the relationship of MTE result significance and the size
of the test suite, etc. For a better clarification of the frame, several
experiment results are analyzed relating human evaluation, BLEU
evaluation, and typological MTE. A visualization method is
introduced for better presentation of the results. The study aims for
aid in construction of test suite and method selection in MTE
practice.
Abstract: In the paper a method of modeling text for Polish is
discussed. The method is aimed at transforming continuous input text
into a text consisting of sentences in so called canonical form, whose
characteristic is, among others, a complete structure as well as no
anaphora or ellipses. The transformation is lossless as to the content
of text being transformed. The modeling method has been worked
out for the needs of the Thetos system, which translates Polish
written texts into the Polish sign language. We believe that the
method can be also used in various applications that deal with the
natural language, e.g. in a text summary generator for Polish.
Abstract: The need for multilingual communication in Japan has
increased due to an increase in the number of foreigners in the
country. When people communicate in their nonnative language,
the differences in language prevent mutual understanding among
the communicating individuals. In the medical field, communication
between the hospital staff and patients is a serious problem. Currently,
medical translators accompany patients to medical care facilities, and
the demand for medical translators is increasing. However, medical
translators cannot necessarily provide support, especially in cases in
which round-the-clock support is required or in case of emergencies.
The medical field has high expectations from information technology.
Hence, a system that supports accurate multilingual communication is
required. Despite recent advances in machine translation technology,
it is very difficult to obtain highly accurate translations. We have
developed a support system called M3 for multilingual medical
reception. M3 provides support functions that aid foreign patients in
the following respects: conversation, questionnaires, reception procedures,
and hospital navigation; it also has a Q&A function. Users
can operate M3 using a touch screen and receive text-based support.
In addition, M3 uses accurate translation tools called parallel texts
to facilitate reliable communication through conversations between
the hospital staff and the patients. However, if there is no parallel
text that expresses what users want to communicate, the users cannot
communicate. In this study, we have developed a circulating support
environment for multilingual medical communication using parallel
texts. The proposed environment can circulate necessary parallel texts
through the following procedure: (1) a user provides feedback about
the necessary parallel texts, following which (2) these parallel texts
are created and evaluated.
Abstract: Machine Translation (MT 3) of English text to its Urdu equivalent is a difficult challenge. Lot of attempts has been made, but a few limited solutions are provided till now. We present a direct approach, using an expert system to translate English text into its equivalent Urdu, using The Unicode Standard, Version 4.0 (ISBN 0-321-18578-1) Range: 0600–06FF. The expert system works with a knowledge base that contains grammatical patterns of English and Urdu, as well as a tense and gender-aware dictionary of Urdu words (with their English equivalents).
Abstract: Named Entity Recognition (NER) aims to classify each word of a document into predefined target named entity classes and is now-a-days considered to be fundamental for many Natural Language Processing (NLP) tasks such as information retrieval, machine translation, information extraction, question answering systems and others. This paper reports about the development of a NER system for Bengali and Hindi using Support Vector Machine (SVM). Though this state of the art machine learning technique has been widely applied to NER in several well-studied languages, the use of this technique to Indian languages (ILs) is very new. The system makes use of the different contextual information of the words along with the variety of features that are helpful in predicting the four different named (NE) classes, such as Person name, Location name, Organization name and Miscellaneous name. We have used the annotated corpora of 122,467 tokens of Bengali and 502,974 tokens of Hindi tagged with the twelve different NE classes 1, defined as part of the IJCNLP-08 NER Shared Task for South and South East Asian Languages (SSEAL) 2. In addition, we have manually annotated 150K wordforms of the Bengali news corpus, developed from the web-archive of a leading Bengali newspaper. We have also developed an unsupervised algorithm in order to generate the lexical context patterns from a part of the unlabeled Bengali news corpus. Lexical patterns have been used as the features of SVM in order to improve the system performance. The NER system has been tested with the gold standard test sets of 35K, and 60K tokens for Bengali, and Hindi, respectively. Evaluation results have demonstrated the recall, precision, and f-score values of 88.61%, 80.12%, and 84.15%, respectively, for Bengali and 80.23%, 74.34%, and 77.17%, respectively, for Hindi. Results show the improvement in the f-score by 5.13% with the use of context patterns. Statistical analysis, ANOVA is also performed to compare the performance of the proposed NER system with that of the existing HMM based system for both the languages.