In Conversation with Eirini Zafeiridou, Welocalize AI Engineer
Studying while working is certainly a challenge; juggling the responsibilities of your job, education, and personal life can be tough to balance. But when you are passionate about the subject, like our Artificial Intelligence (AI) and Machine Learning Engineer, Eirini Zafeiridou, it’s definitely worth it.
Based in Greece, Eirini is currently studying for an M.S.c in Language Technology while working at Welocalize. We chatted with Eirini to learn more about her Master’s research within language technology, her role at Welocalize, how she finds working while studying, and where she thinks the future of language technology is heading.
Hi, Eirini. What is your role at Welocalize? What does a typical day look like for you?
I am a member of Welocalize’s artificial intelligence (AI Research and Development (R&D) team. I mainly work within machine translation (MT) and the broader natural language processing (NLP) field. I am responsible for the implementation of end-to-end NLP projects, including but not limited to Neural Machine Translation (NMT), Quality Estimation, Text Generation, Text Mining, Processing and Classification, Information Retrieval, as well as Terminology Extraction. On a daily basis, I research, experiment with, and develop state-of-the-art NLP solutions to deal with complex business problems and objectives.
How did you get started in the language technology/MT space?
With a background in Language and Linguistics, I was always fascinated by the way that the human brain acquires, develops, processes, and produces natural language. Expanding this to cover AI as well, I was intrigued to learn how computers analyze and generate large amounts of language data. NLP is an interdisciplinary field concerned with the interaction between human language and computers, answering in this way a lot of my early-career scientific questions. After having successfully completed several undergraduate courses in Python and Computational Linguistics, I started working in the Conversational AI space as a Computational Linguist being mainly responsible for building chatbots and spoken dialogue systems for a wide range of international telecommunication and banking clients. By becoming increasingly more active in the NLP industry, I made the decision to proceed with postgraduate studies in NLP.
Tell us more about your Master’s degree – what are you studying?
I am currently studying for an M.Sc. in “Language Technology”, a postgraduate degree co-organized by the National and Kapodistrian University of Athens and the “Athena” Research Center. The main objective of this Master’s degree is to prepare and educate scientists in the field of NLP as well as to promote research and highlight recent technological advancements.
The degree covers a wide range of NLP applications such as Machine Translation (MT), Deep Learning, Big Data Management, Human-Computer Interaction, NLP, Natural Language Understanding (NLU), Natural Language Generation (NLG), Machine Learning, and so much more!
I am now in the process of building my final thesis titled “Quality Estimation for Neural Machine Translation” which resulted in a shared task participation and a paper submission to the EMNLP 2022 & WMT Seventh Conference on MT.
That’s exciting! Your paper was accepted, and you recently presented at EMNLP 2022 & WMT Seventh Conference on MT, a leading conference in the area of NLP & AI in Abu Dhabi – can you share more about the event and your presentation?
As part of my M.Sc thesis, I built a deep learning model that automatically estimates the quality of the Machine Translation output without the need for a reference translation or any other human input. More specifically, with the model I developed, we participated in the Sentence-level Quality Prediction subtask of the WMT 2022 Shared Task on Quality Estimation. Our system exhibited competitive results outperforming the Baseline in terms of both Spearman and Pearson correlation coefficient. Our contribution resulted in a scientific paper that was accepted and presented at EMNLP 2022, a leading conference in the area of NLP & AI.
How does your Master’s benefit your role at Welocalize?
My Master’s degree is highly related to my current role at Welocalize. It has helped me acquire additional technical skills necessary for my everyday working activities by providing me with structured knowledge, advanced education, research methodologies, and a deeper understanding of the Machine Learning and NLP fields. My studies also allow me to enhance my soft skills by learning to study independently under pressure and be well-organized, self-directed, and self-driven. I manage to set explicit goals, clarify, and efficiently implement my ideas, focus on my efforts, multitask, prioritize, and use my time productively.
How does your research, ‘Quality Estimation in NMT,’ benefit the language industry as a whole?
In our research study, we suggested the use of external features which demonstrates a strong correlation with human judgments of quality assessment. This particular set of features, which represents various levels of translation quality, was extracted from pre-trained models and was used during the training of our system. We believe that our participation was a significant contribution to the WMT 2022 and more specifically to the Quality Estimation shared task.
What advice do you have for others who are looking to study for a Master’s, while working?
Combining a job, postgraduate studies, and personal life is quite difficult but not impossible. To properly achieve a balance between those, one must make considerable time and effort sacrifices. High motivation, devotion, and enthusiasm for the taught subject are definitely the key to overcoming such challenges.
Where do you see the future of language technology heading? Is there anything you’re excited to see develop in the future?
Speaking with a computer or automatically translating a document seemed unrealistic several years ago. NLP is now growing in an exponential manner, and it has become a crucial part of our everyday lives more than ever before. Nowadays, we do have the ability to communicate with virtual assistants, automate several daily tasks, machine translate text from one language to another, summarize documents, discover customer analytics, or even automatically search and get recommendations using the available search engines.
However, there are certain areas where NLP still lacks. I see the future in a more humanized version of AI which may respond to the user in a way that is more human-like or natural by clearly understanding pragmatics, rhetorical devices such as metaphorical use of language, contextual information, cultural aspects, non-verbal communication, or even human emotions when humans express them through voice, facial expressions, or physical movements.
Further Reading You’ll Love…
For more information on Welocalize language and translation AI and technology solutions, connect with us here.