Terminology Alignment faces big challenges in NLP because of the dynamic nature of terms. Fortunately, over these last few years, Deep Learning models showed very good progress with several NLP tasks such as multilingual data resourcing, glossary building, terminology understanding. . . etc. In this work, we propose a new method for terminology alignment from a comparable corpus (Arabic/French languages) for the Algerian culture field. We aim to improve bilingual alignment based on contextual information of a term and to create a significant term bank i.e. a bilingual Arabic-French dictionary. We propose to create word embeddings for both Arabic and French languages using ELMO model focusing on contextual features of terms. Then, we mapp those embeddings using Seq2seq model. We use multilingual-BERT and All-MiniLM-L6 as baseline mod- els to compare terminology alignment results. Lastly we study the performance of these models by applying evaluation methods. Experimentation’s showed quite satisfying alignment results.

We propose a tool for the semi-automatic production of terminological databases, divided in the steps of corpus processing, terminology extraction, database population and management. With this tool it is possible to obtain a draft macrostructure (a lemma-list) and data for the microstructural level, such as grammatical (morphosyntactic patterns, gender, formation process) and semantic information (hypernyms, equivalence in another language, definitions and synonyms). In this paper we offer an overall description of the software and an evaluation of its performance, for which we used a linguistics corpus in English and Spanish.

Use of NLP Techniques in Translation by ChatGPT: Case Study Natural Language Processing (NLP) refers to a field of study within the domain of artificial intelligence (AI) and computational linguistics that focuses on the interaction between computers and human language. NLP seeks to develop computational models and algorithms capable of understanding, analyzing, and generating natural language text and speech (Brown et al., 1990). At its core, NLP aims to bridge the gap between human language and machine understanding by employing various techniques from linguistics, computer science, and statistics. It involves the application of linguistic and computational theories to process, interpret, and extract meaningful information from unstructured textual data (Bahdanau, Cho and Bengio, 2015). Researchers and practitioners in NLP employ diverse methodologies, including rule-based approaches, statistical models, machine learning techniques (such as neural networks), and more recently, deep learning architectures. These methodologies enable the development of robust algorithms that can learn from large-scale language data to improve the accuracy and effectiveness of language processing systems (Nilsson, 2010). NLP has numerous real-world applications across various domains, including information retrieval, virtual assistants, chatbots, social media analysis, sentiment monitoring, automated translation services, and healthcare, among others (kaynak). As the field continues to advance, NLP strives to overcome challenges such as understanding the nuances of human language, handling ambiguity, context sensitivity, and incorporating knowledge from diverse sources to enable machines to effectively communicate and interact with humans in a more natural and intuitive manner. Natural Language Processing (NLP) and translation are interconnected fields that share a symbiotic relationship, as NLP techniques and methodologies greatly contribute to the advancement and effectiveness of machine translation systems. NLP, a subfield of artificial intelligence (AI), focuses on the interaction between computers and human language. It encompasses a wide range of tasks, including text analysis, syntactic and semantic parsing, sentiment analysis, information extraction, and machine translation (Bahdanau, Cho and Bengio, 2014). NMT models employ deep learning architectures, such as recurrent neural networks (RNNs) and more specifically, long short-term memory (LSTM) networks, to learn the mapping between source and target language sentences. These models are trained on large-scale parallel corpora, consisting of aligned sentence pairs in different languages. The training process involves optimizing model parameters to minimize the discrepancy between predicted translations and human-generated translations (Wu et al., 2016) NLP techniques are crucial at various stages of machine translation. Preprocessing techniques, such as tokenization, sentence segmentation, and morphological analysis, help break down input text into meaningful linguistic units, making it easier for translation models to process and understand the content. Syntactic and semantic parsing techniques aid in capturing the structural and semantic relationships within sentences, improving the overall coherence and accuracy of translations. Furthermore, NLP-based methods are employed for handling specific translation challenges, such as handling idiomatic expressions, resolving lexical ambiguities, and addressing syntactic divergences between languages. For instance, statistical alignment models, based on NLP algorithms, enable the identification of correspondences between words or phrases in source and target languages, facilitating the generation of more accurate translations (kaynak). Several studies have demonstrated the effectiveness of NLP techniques in enhancing machine translation quality. For example, Bahdanau et al. (2015) introduced the attention mechanism, an NLP technique that enables NMT models to focus on relevant parts of the source sentence during translation. This attention mechanism significantly improved the translation quality of neural machine translation models. ChatGPT is a language model developed by OpenAI that utilizes the principles of Natural Language Processing (NLP) for various tasks, including translations. NLP is a field of artificial intelligence that focuses on the interaction between computers and human language. It encompasses a range of techniques and algorithms for processing, analyzing, and understanding natural language. When it comes to translation, NLP techniques can be applied to facilitate the conversion of text from one language to another. ChatGPT employs a sequence-to-sequence model, a type of neural network architecture commonly used in machine translation tasks. This model takes an input sequence in one language and generates a corresponding output sequence in the target language (OpenAI, 2023). The training process for ChatGPT involves exposing the model to large amounts of multilingual data, allowing it to learn patterns, syntax, and semantic relationships across different languages. This exposure enables the model to develop a general understanding of language structures and meanings, making it capable of performing translation tasks. To enhance translation quality, ChatGPT leverages the Transformer architecture, which has been highly successful in NLP tasks. Transformers utilize attention mechanisms, enabling the model to focus on different parts of the input sequence during the translation process. This attention mechanism allows the model to capture long-range dependencies and improve the overall coherence and accuracy of translations. Additionally, techniques such as subword tokenization, which divides words into smaller units, are commonly employed in NLP translation systems like ChatGPT. Subword tokenization helps handle out-of-vocabulary words and improves the model’s ability to handle rare or unknown words (GPT-4 Technical Report, 2023). As can be seen, there have been significant developments in artificial intelligence translations thanks to NLP. However, it is not possible to say that it has fully reached the quality of translation made by people. The only goal in artificial intelligence translations is to reach translations made by humans. In general, there are some fundamental differences between human and ChatGPT translations. Human-made translations and translations generated by ChatGPT (or similar language models) have several key differences (Kelly and Zetzsche, 2014; Koehn, 2010; Sutskever, Vinyals and Le, 2014; Costa-jussà and Fonollosa, 2018) Translation Quality: Human translators are capable of producing high-quality translations with a deep understanding of both the source and target languages. They can accurately capture the nuances, cultural references, idioms, and context of the original text. On the other hand, ChatGPT translations can sometimes be less accurate or may not fully grasp the intended meaning due to the limitations of the training data and the model’s inability to comprehend context in the same way a human can. While ChatGPT can provide reasonable translations, they may lack the finesse and precision of a human translator. Natural Language Processing: Human translators are skilled at processing and understanding natural language, taking into account the broader context, cultural implications, and the intended audience. They can adapt their translations to suit the target audience, tone, and purpose of the text. ChatGPT, although trained on a vast amount of text data, lacks the same level of natural language understanding. It often relies on pattern matching and statistical analysis to generate translations, which can result in less nuanced or contextually appropriate outputs. Subject Matter Expertise: Human translators often specialize in specific domains or subject areas, allowing them to have deep knowledge and understanding of technical or specialized terminology. They can accurately translate complex or industry-specific texts, ensuring the meaning is preserved. ChatGPT, while having access to a wide range of general knowledge, may struggle with domain-specific vocabulary or terminology, leading to inaccuracies or incorrect translations in specialized texts. Cultural Sensitivity: Human translators are well-versed in the cultural nuances of both the source and target languages. They can navigate potential pitfalls, adapt the translation to the cultural context, and avoid unintended offensive or inappropriate language choices. ChatGPT lacks this level of cultural sensitivity and may produce translations that are culturally tone-deaf or insensitive, as it lacks the ability to understand the subtleties and implications of language choices. Revision and Editing: Human translators go through an iterative process of revision and editing to refine their translations, ensuring accuracy, clarity, and quality. They can self-correct errors and refine their translations based on feedback or additional research. ChatGPT, while capable of generating translations, does not have the same ability to self-correct or improve based on feedback. It generates translations in a single pass, without the iterative refinement process that humans can employ. In summary, while ChatGPT can be a useful tool for generating translations, human-made translations generally outperform machine-generated translations in terms of quality, accuracy, contextuality, cultural sensitivity, and domain-specific expertise. In conclusion, NLP and machine translation are closely intertwined, with NLP providing essential tools, methodologies, and techniques that contribute to the development and improvement of machine translation systems. The integration of NLP methods has led to significant advancements in translation accuracy, fluency, and the ability to handle various linguistic complexities. As NLP continues to evolve, its impact on the field of machine translation is expected to grow, enabling the creation of more sophisticated and context-aware translation systems. On the basis of all this information, in this research, it is aimed to compare the translations from English to Turkish made by ChatGPT, one of the most advanced artificial intelligences, with the translations made by humans. In this context, an academic 1 page English text was chosen. The text was translated by both ChatGPT and a translator who is an academic in the field of translation and has 10 years of experience. Afterwards, two different translations were examined comparatively by 5 different translators who are experts in their fields. Semi-structured in-depth interviews were conducted with these translators. The aim of this study is to reveal the role of artificial intelligence tools in translation, which are increasing day by day and suggesting that there will be no need for language learning in the future. On the other hand, many translators argue that artificial intelligence and human translations can be understood. Therefore, if artificial intelligence is successful, there will be no profession called translator in the future. This research seems to be very useful in terms of shedding light on the future. The method of this research is semi-structured in-depth interview. References Bahdanau, D., Cho, K. and Bengio Y. (2015). Neural machine translation by jointly learning to align and translate. In International Conference on Learning Representations. Brown, P. F., Cocke, J., Pietra, S. A. D., Pietra, V. J. D., Jelinek, F., Lafferty, J. D., Mercer, R. L., and Roossin, P. S. A. (1990) statistical approach to machine translation. Computational linguistics 16, 2, 79–85. Costa-jussà, M. R., & Fonollosa, J. A. R. (2018). “An Overview of Neural Machine Translation.” IEEE Transactions on Neural Networks and Learning Systems. GPT-4 Technical Report (2023). https://arxiv.org/abs/2303.08774. Kelly, N. and Zetzsche, J. (2014). Found in Translation: How Language Shapes Our Lives and Transforms the World. USA: Penguin Book. Koehn, P. (2010). “Statistical Machine Translation.” Cambridge University Press. Nilsson, N. J. (2010). The Quest For AI- A History Of Ideas And Achievements. http://ai.standford.edu/ nilsson/. OpenAI (2023). https://openai.com/blog/chatgpt/. Sutskever, I., Vinyals, O., & Le, Q. V. (2014). “Sequence to Sequence Learning with Neural Networks.” Advances in Neural Information Processing Systems. Wu,Y. Schuster, M., Chen, Z., Le, Q. V. and Norouzi M. (2016). Google’s Neural Machine Translation System: Bridging the Gap between Human and Machine Translation. https://arxiv.org/pdf/1609.08144.pdf.

In the translation process, terminological resources are used to solve translation problems, so information on terminological equivalence is crucial to make the most appropriate choices in terms of translation equivalence. In the context of Machine translation, indeed, neural models have improved the state-of-the-art in Machine Translation considerably in recent years. However, they still underperform in domain-specific fields and in under-resourced languages. This is particularly evident in translating legal terminology for Arabic, where current Machine Translation outputs do not adhere to the contextual, linguistic, cultural, and terminological constraints posed by translating legal terms in Arabic. In this paper, we conduct a comparative qualitative evaluation and comprehensive error analysis on legal terminology translation in Phrase-Based Statistical Machine Translation and Neural Machine Translation in two translation language pairs: Arabic-English and Arabic-French. We propose an error typology taking the legal terminology translation from Arabic into account. We demonstrate our findings, highlighting the strengths and weaknesses of both approaches in the area of legal terminology translation for Arabic. We also introduce a multilingual gold standard dataset that we developed using our Arabic legal corpus. This dataset serves as a reliable benchmark and/or reference during the evaluation process to decide the degree of adequacy and fluency of the Phrase-Based Statistical Machine Translation and Neural Machine Translation systems.

Implementing technology in a modern-day classroom is an ongoing challenge. In this paper, we created a system for an automatic assessment of student answers using Latent Semantic Analysis (LSA) – a method with an underlying assumption that words with similar meanings will appear in the same contexts. The system will be used within digital lexical flash-cards for L2 vocabulary acquisition in a CLIL classroom. Results presented in this paper indicate that while LSA does well in creating semantic spaces for longer texts, it somewhat struggles with detecting topics in short texts. After obtaining LSA semantic spaces, answer accuracy was assessed by calculating the cosine similarity between a student’s answer and the golden standard. The answers were classified by accuracy using KNN, for both binary and multinomial classification. The results of KNN classification are as follows: precision P = 0.73, recall R = 1.00, F1 = 0.85 for binary classification, and P = 0.50, R = 0.47, F1 = 0.46 score for the multinomial classifier. The results are to be taken with a grain of salt, due to a small test and training dataset.

This paper represents a description of Bulgarian verbal computer terms with a view to the specifics of their translation in English. The study employs a subset of 100 verbs extracted from the Bulgarian WordNet (BulNet) and from the internet. The analysis of their syntactic and semantic structure is a part of a study of the general lexis of Bulgarian. The aim of the paper is to (1) identify some problem areas of the description and translation of general lexis verbs, (2) offer an approach to the semantic description of metaphor-based terms from the perspective of Frame Semantics; (3) raise questions about the definition of general lexis with respect to Bulgarian and across languages.

Initial work has been done to address fake news detection and misrepresentation of news in the Bengali language. However, no work in Bengali yet addresses the identification of specific claims in social media news that falsely manipulate a related news article. At this point, this problem has been tackled in English and a few other languages, but not in the Bengali language. In this paper, we curate a dataset of social media content labeled with information manipulation relative to reference articles, called BanMANI. The dataset collection method we describe works around the limitations of the available NLP tools in Bangla. We expect these techniques will carry over to building similar datasets in other low-resource languages. BanMANI forms the basis both for evaluating the capabilities of existing NLP systems and for training or fine-tuning new models specifically on this task. In our analysis, we find that this task challenges current LLMs both under zero-shot and fine-tuned set- things

This study, submitted to the BUCC2023 shared task on bilingual term alignment in comparable specialised corpora, introduces a supervised, feature-based classification approach. The approach employs both static cross-lingual embeddings and contextual multilingual embeddings, combined with surface-level indicators such as Levenshtein distance and term length, as well as linguistic information. Results exhibit improved performance over previous methodologies, illustrating the merit of integrating diverse features. However, the error analysis also reveals remaining challenges.