Machine Translation Research Articles

Accuracy of Google Translate in Translation of English-Kiswahili and Kiswahili-English Newspaper Headlines

This paper endeavors to determine the accuracy of Google Translate in newspaper headlines from Kiswahili to English and vice versa, while using the human translator as the yardstick. Newspaper headlines in both Kiswahili and English were identified and randomly selected. Three human translators were used to so that the Google Translate translations could be measured against the human translators. The Relevance Theory was applied during the research. This study made use of both Qualitative and Quantitative Research Methodology and a Descriptive Research Design. Simple Random Sampling was used to select the data to be used while Purposive Sampling was used when choosing the human translators. Fifty data sets were tested, twenty-five of which were in Kiswahili while the rest were in English. Content analysis was thereafter applied to interpret the translation output. The study found that the human translator is more accurate than Google Translate. In addition, some human translations were found to slightly differ from Google translations in the wording but still had the same meaning. The study focused on the communicativeness of the translated data and found that some items translated exhibited meaning losses. It was found that Google Translate was able to accurately render the meaning of 28/50 (56%) of the instructions examined, implying that it is 56% accurate in translating Kiswahili to English and vice versa. Mistranslations were found to be more prevalent in the Kiswahili source data. This therefore means that sometimes miscommunication occurs as some items are not accurately rendered. This study thus offers useful insight on areas of intervention in Machine Translation, particularly Google Translate.

Self-Supervised Multimodal Learning: A Survey.

Multimodal learning, which aims to understand and analyze information from multiple modalities, has achieved substantial progress in the supervised regime in recent years. However, the heavy dependence on data paired with expensive human annotations impedes scaling up models. Meanwhile, given the availability of large-scale unannotated data in the wild, self-supervised learning has become an attractive strategy to alleviate the annotation bottleneck. Building on these two directions, self-supervised multimodal learning (SSML) provides ways to learn from raw multimodal data. In this survey, we provide a comprehensive review of the state-of-the-art in SSML, in which we elucidate three major challenges intrinsic to self-supervised learning with multimodal data: (1) learning representations from multimodal data without labels, (2) fusion of different modalities, and (3) learning with unaligned data. We then detail existing solutions to these challenges. Specifically, we consider (1) objectives for learning from multimodal unlabeled data via self-supervision, (2) model architectures from the perspective of different multimodal fusion strategies, and (3) pair-free learning strategies for coarse-grained and fine-grained alignment. We also review real-world applications of SSML algorithms in diverse fields such as healthcare, remote sensing, and machine translation. Finally, we discuss challenges and future directions for SSML. A collection of related resources can be found at: https://github.com/ys-zong/awesome-self-supervised-multimodal-learning.

Design of Translation Error Correction System Based on Improved Seq2Seq

Speech translation technology is an important booster for promoting social communication and advancing human civilization. With the solid advancement of theories and technologies such as speech processing and machine translation, as well as the continuous deepening and development of computer science, the computing power and storage capacity have been further improved. Speech translation systems widely used in English, French, English and Chinese have successively reached a commercial level. However, the development of speech translation systems is limited by corpus resources and a lack of bilingual language research. The experiments and applications of speech translation in some languages are still in its early stages. In addition, existing research mostly adopts a cascading voice translation system as the basis, breaking through the key issues individually, and less adopts direct voice translation methods without relying on intermediate text representation, which brings problems such as cumbersome research content, complex translation models, and long translation delays. This article mainly focuses on the design of a translation error correction system based on improved Seq2Seq. Firstly, relevant research is summarized, relevant application methods are proposed, and the results are discussed. I hope to bring some inspiration to relevant researchers and provide assistance for the development of related fields.

Construction of Chinese Culture Teaching Resources for International Chinese Language Education Based on Knowledge Mapping

Abstract The transformative development of international Chinese language education has entered a new period, showing emerging development trends such as diversification, marketization, and digitalization. This study creates a unified classification system that links Chinese and English knowledge and complements each other based on the cross-language knowledge mapping of Chinese cultural knowledge. Additionally, a knowledge base for sharing cultural knowledge between Chinese and English is being constructed to provide strong support for information retrieval and machine translation. The data information is stored in plain text format for ontology and knowledge to make it easier to use and process teaching resources. The results show that in the teaching feedback, the average scores are all over 3, indicating that the teaching effect of the resource is positive overall, and the standard deviation values range from 0.1560 to 0.9812, indicating significant differences between different individuals. The above findings can be a valuable resource for constructing and implementing international Chinese language education network resources.

Lexicon-based Language Relatedness Analysis

The field of computational linguistics has been impacting various issues in language disciplines. The enormous growth of machine learning algorithms and Natural Language Processing (NLP) empowers its advancement and brings huge benefits to societies. For instance, machine translation, text summarization, sentence auto-completion, and sentiment analysis are a few of its benefits. However, leveraging this opportunity for low-resourced languages is challenging due to the lack of available electronic datasets. This paper presents a lexicon-based language relatedness analysis on Ethiopian low-resourced languages. The languages Wolaita, Dawuro, Gamo, and Gofa belong to the Ethiopian Omotic language family and share rich linguistic cultures and similarities. However, the extent of their inter-relatedness remains unknown. To address this gap, we collected and prepared novel corpora from the Bible and academic texts. We employed the TF-IDF technique for feature extraction and used the cosine similarity method to measure the similarities among these languages. In addition to cosine similarity, we used Euclidean distance to measure the spatial distances between the languages. The experiment results showed that Wolaita and Gofa exhibited high relatedness (33.4%), while Dawuro and Gamo demonstrated low relatedness (12.1%).

The Promise and Pitfalls of AI Technology in Bridging Digital Language Divide: Insights from Machine Translation on Wikipedia

The Promise and Pitfalls of AI Technology in Bridging Digital Language Divide: Insights from Machine Translation on Wikipedia

Arabic NER Evaluation: Pre-Trained Models via Contrastive Learning vs. LLM Few-Shot Prompting

Developing Natural Language Processing (NLP) tools for the Arabic language and its dialects is very challenging. Named Entity Recognition (NER) is one of these challenges, which serves as the core component in many NLP systems such as information extraction, question answering, machine translation and knowledge graph building. This paper sheds light on applying diferent approaches for Arabic NER (Flat and Nested) using a large and rich Arabic NER corpus, Wojood dataset, which consists of about 550K tokens annotated with 21 entity types. First, we apply the Wojood base model, AraBERTv2, along with various other Arabic BERT models such as MARBERTv2, CaMelBert, mBert, ..etc. Next, we utilize the Bi-Encoder Contrastive Learning (CL) approach, a framework developed by Microsoft, which maps candidate text spans and entity types into the same vector representation space. The primary challenge in this approach is distinguishing non-entity spans from entity mentions. This approach could achieve F1 score 91.25% for Flat and 91.40% for Nested NER. Additionally, for evaluating the predicted NER, we employ Few-Shot prompting on LLaMA, and GPT-3.5 using refined prompt-based strategy. Our findings reveal that LLaMA outperforms GPT3.5.

A Review On Semantic Role Labelling For Indian Languages

Semantics means the process of extracting precise meaning from text or sentences. Semantic Role Labelling (SRL) plays a crucial role in Natural Language Processing by providing insights into the underlying meaning of sentences. SRL involves assigning generic labels or roles to words within a sentence, indicating their respective semantic roles. It helps in tasks such as information extraction, question answering, educational systems, sentiment analysis, and machine translation by identifying the roles of different words in a sentence and their relationships with each other. This process enables computers to better understand and process human language, leading to more accurate and effective language understanding. SRL also powers AI-driven educational tools like intelligent tutoring systems, personalized learning, and automated assessments, making education more adaptive and effective. In this paper, we are giving a brief review of SRL system developed for different languages. This literature review focuses on key aspects of SRL like techniques used, datasets used, languages for which SRL is developed, and accuracy. Our future work is inclined towards SRL for Hindi, so our review focuses on SRL developed for Hindi, along with other Indian Languages like Urdu, Malayalam and Tamil.

Technology for Public Service Translators and Interpreters in Spain: Enhancing Employability Through Training

Despite rapid technological advances and how they affect public service interpreting and translation (psit), as a subfield of translation and interpretation, very few studies focus on how technological competence affects or facilitates employability in psit. This descriptive study carried out among graduates of the University of Alcalá’s psit programme intends to fill that gap by examining the usefulness of its technology-related training. This descriptive case study sought (1) to establish which technological tools graduates had to use in their jobs and the technological requirements in the labour market for psit; (2) to establish potential differences between the T&I and the psit sectors regarding technological requirements or needs; and (3) to assess how useful the training received was. Findings show similarities in the use of sources and tools in both T&I and psit, but also differences in information mining sources, highlighting varied needs. Moreover, although computer-assisted-translation tools and machine translation are widely demanded nowadays, their use by graduates was shown to be lower than expected. This study gives critical insights for researchers, trainers, and programme designers to ensure that their syllabi encompass comprehensive content tailored to both in-house and freelance translators to increase their employability.

Deep Learning based Named Entity Recognition for the Bodo Language

One of the important application of natural language processing (NLP) is Name Entity Recognition (NER). It automatically recognise and categorise named entities in a document. Named Entities can be the name of an individual, group, place, etc. It is crucial to the success of NER applications including text summarization, machine translation and information extraction and retrieval. It is one of the most useful application tools for a variety of topics and languages. Despite its widespread use and effectiveness in English, this field is currently under investigation for other Indian languages, such as Bodo. Due to the lack of resources and a high-quality dataset, NER in Bodo is a difficult task. In this research, a deep learning-based NER tagger is investigated for the Bodo language and NER tagged dataset is generated for Bodo language using Docanno and enlarge the dataset size by employing a data augmentation technique. As there is no Bodo NER baseline model to compared with, we employed several deep learning techniques for Bodo NER System and compared their results. We achieved an accuracy of 99.62%, precision of 99.75%, recall of 98.74% and F-score of 99.35% when employed with LSTM and character based. This study also highlights GRU and CNN based models performance in Bodo NER task.

Latent Semantic and Disentangled Attention.

Sequential learning using transformer has achieved state-of-the-art performance in natural language tasks and many others. The key to this success is the multi-head self attention which encodes and gathers the features from individual tokens of an input sequence. The mapping or decoding is performed to produce an output sequence via cross attention. There are threefold weaknesses by using such an attention framework. First, since the attention would mix up the features of different tokens in input and output sequences, it is likely that redundant information exists in sequence data representation. Second, the patterns of attention weights among different heads tend to be similar. The model capacity is bounded. Third, the robustness in an encoder-decoder network against the model uncertainty is disregarded. To handle these weaknesses, this paper presents a Bayesian semantic and disentangled mask attention to learn latent disentanglement in multi-head attention where the redundant features in transformer are compensated with the latent topic information. The attention weights are filtered by a mask which is optimized through semantic clustering. This attention mechanism is implemented according to Bayesian learning for clustered disentanglement. The experiments on machine translation and speech recognition show the merit of Bayesian clustered disentanglement for mask attention.

Research on the Application of Natural Language Processing Technology in English Translation Quality Control

Abstract Machine translation is an important research area in natural language processing. In this paper, we propose an innovative fusion translation model that combines the slicing method and the alignment method and select articles from different domains for experiments to explore the translation effects of the proposed method. The experiments show that the proposed CRF sequence annotation cut-scoring method performs excellently, with F-values of 86.7%, 89.0%, and 87.4% for the three domains of computer science, civil engineering, and medicine, respectively. The hybrid alignment method based on the length and lexical information also performs well, both in terms of correctness and recall and the hybrid alignment method is able to obtain better results than the length or lexical information methods alone. Putting the proposed method into the translation accuracy test, the BLEU score of the cut-and-align fusion translation model is improved to 11.08, while the NIST score is improved to 3.6468, which is a significant improvement in the translation quality of English.

SOFTWARE IMPLEMENTATION OF AUTOMATED USER INTERFACE TRANSLATION TOOL

Numerous daily activities are long since accomplished using mobile applications and international resources available through the Internet. The majority of such resources support only a limited amount of world languages for the interface. This raises the issue of the need for end users to know the languages required to operate and use these programs. Not every developer in the world has access to professional translation services, or the ability to create such translations on their own. Evaluating these factors shows an existing need of a convenient and accessible program specialized in creating and ensuring quality translation of software user interfaces specifically. Analysis was conducted to determine advantages and disadvantages of possible implementation methods and existing algorithms, i.e. GNU gettext. Based on this analysis requirements for the resulting program are formed, those being a comfortable user interface and a set of functions specific to working with user interfaces and source code files. The list of primary functionality includes interaction with internet services for machine translation and the ability to view the source code of a program or website in context of text elements of the user interface. A program tool for the Microsoft Windows environment that implements all requirements to base functionality is conceptualized and developed. The project version of the program was implemented using C# programming language and Microsoft Visual Studio development environment. The program is specialized for the translation of user interfaces, and enables automation of this process via usage of the Google Translate service. To maximize ease of use the program stores the resulting translation in a simple text format, designed to be loaded and read from during runtime of the target program or website. Also included is a detailed demonstration on the workflow using the interface of the translation program as an example, and potential directions of further developments are determined. Keywords: machine translation; software development; user interface; Google Translate

A Method for Extracting BPMN Models from Textual Descriptions Using Natural Language Processing

Business Process Model and Notation (BPMN) is a standard for formally modeling complex business processes. Manual creation of BPMN models can be time-consuming and error-prone, prompting a need for automation. Existing approaches, such as rule-based methods, machine learning, and machine translation, have progressed but face accuracy and real-world applicability challenges. In this research paper, we propose a novel method for automated extraction of BPMN models from textual descriptions using natural language processing (NLP) tools and deep learning models, including the spaCy library for text processing, a fine-tuned BERT model, and state-of-the-art large language models like GPT-3.5-Turbo and GPT-4. We utilize Graphviz, an open-source graph visualization software, to visualize the extracted processes. Our method supports representing tasks, exclusive gateways, parallel gateways, and start and end events in the generated BPMN models. The evaluation of 31 textual descriptions shows that our method generates process models with 96% accuracy using GPT-4 and 80% accuracy using GPT-3.5-Turbo large language models. Although subject to certain limitations, such as occasional inaccuracies in model outputs and reliance on well-formed input text, our approach offers a valuable contribution to the growing body of research on automating BPMN model generation.

The interpretative approach as a platform for forming would-be translators’ professional intelligence

The article deals with theoretical and practical issues of forming would-be translators and interpreters’ professional intelligence. The concept of a translator’s professional intelligence is regarded as a structured element of cognitive skills. A person’s intelligence is a mental process interpreted by P. Guilford as an ability of processing information. A translator’s professional intelligence uses mental and physiological devices such as memory, attention, imagination, reflection etc. and functions as a processor responsible for perceiving, comprehending and deverbalizing a source message. To form the above cognitive skill we have selected the interoretative theory of translation pioneered by French scholars D. Seleskovicz and M. Lederer. This model involves a transition between two one-language communication acts and rendering the sense of the source message without focusing on its linguistic form. We believe that this approach is the best to demonstrate the concept of translation between languages (senses) and alienates it from instructions put into translation machines. Our organizational and didactic model is to use a pre-translation stage to form the operations performed by a translator’s professional intelligence (perception, comprehension, procession, and understanding). It is the period of time when a translator is focused on reading and analyzing actively the source message and preparing for generating a new text in a different language. The intellectual processing of the source discourse is due to produce a mental substance to be transcoded and reverbalized. That is what a translator’s professional intelligence is all about.

Increasing United States College Access for Native Arabic Speakers: Applying a Simplification Intervention and Evaluating Machine and Human Translations

Increasing United States College Access for Native Arabic Speakers: Applying a Simplification Intervention and Evaluating Machine and Human Translations

Optimization Strategies and Efficacy Evaluation of Cross-Language Embedding Model in Teaching English in Colleges and Universities

Abstract With the trend of information technology, contemporary English language teaching is moving positively in the direction of intelligence. This study utilizes the text embedding model to represent language and convert it into a format that computers can process. Through the methods of adversarial training and self-learning training and adding the fine-tuning process to improve the effect of cross-linguistic word vectors, the machine reading comprehension and translation model is constructed to ensure the semantic consistency between it and the input source text. Meanwhile, this study combines the model to implement an innovative teaching model of English in colleges and universities and evaluates its effectiveness. The results show that the effectiveness of the proposed method in this paper on this indicator has been narrowed to a gap of 11.48%, which significantly confirms the effectiveness of the proposed model. In the direction of E2C, the cross-linguistic embedding model has the highest average F1 score (0.894). The average score of class 1 results is 82.6843, corresponding to a critical confidence level of 0.036<0.05, which indicates that there is a significant difference between the results of class 1 and class 2. The reference value of this study is important for English teaching and related system development in colleges and universities.

On Applying Natural Language Processing Technology to Optimize Accuracy of English Interactive Translation Platform

Abstract This research establishes a phrase-based statistical machine translation platform composed of multiple functional modules, including data preprocessing, word alignment, translation model training, ordering model training, language model training, decoding, and parameter optimization. Special emphasis is placed on the quality of bilingual word alignment, utilizing both the IBM model and the statistical collocation model to enhance alignment accuracy. The study reveals that this new platform significantly enhances translation efficiency across diverse corpora, such as legal documents, news articles, and parliamentary records. Notably, the translation of parliamentary records demonstrates a remarkable reduction in average translation time by 27.53% and a decrease in the number of decoding iterations by 56.46% when compared to System1 using the Baseline system. Furthermore, a BLEU-based evaluation indicates that the Baseline system consistently surpasses the comparison system in translating various text types. This study underscores the practical effectiveness of the developed translation platform and its contribution to the advancement of natural language processing technology.

Improving the Punjabi-Hindi Braille Neural Machine Translation through Syntax Augmentation

The paper describes the development of a neural-based machine translation system for the Punjabi-to-HindiBraille language pair. As no system can directly (on the fly) translate Punjabi (Gurmukhi) text into Bharati Braille, we worked on developing the same. The system was trained on 6 lac parallel corpus. The training corpus of the system incorporated linguistic information along with the parallel text. The system was compared with the baseline NMT system with the syntax-augmented NMT system. The trained system was tested using the standard BLEU MT evaluation metric. The baseline NMT system scored 26.34 while the syntax-augmented NMT system scored 36.66 which was significantly higher than the baseline system.

Translators’ Resource Dominance and the Success of Finding the Target Terms in Human Translation and Post-editing of Machine Translation

Translators’ Resource Dominance and the Success of Finding the Target Terms in Human Translation and Post-editing of Machine Translation