EBMT / NLP Laboratory
This Laboratory
The Laboratory for Example-Based Machine Translation and Natural Language Processing (EBMT/NLP lab) is a laboratory of the Graduate School of Information, Production and Systems (IPS), Waseda University.
Historically, the lab has been focusing on machine translation, concentrating on analogy-based methods and low-resourced languages (case-based reasoning). In modern terms, our scientific question is to identify regularity from scarce data and learn how to generalize safely and faster from it. We used to apply symbolic methods, e.g., in formal morphology, or in the syntax of short sentences. We studied semantic analogies between words in word embedding models. With the advent of large language models (LLMs), our interest shifted to numerical methods and longer sentences or texts. We study how much linguistic knowledge LLMs capture by inspecting their linguistic competence through analogy questions. We aim at exploiting the analogical structure of language to attempt faster generalization, advocating for the use and development of smaller language models.
Research Topics
- Formalization of analogy: analogy between strings, numerical analogy
- Creation of analogy datasets, for any language data
- Linguistic competence of large and smaller language models
- Analogical reasoning in classical and generative AI
- Development of analogy-based neural networks
External Links
Contact
Graduate School of Information,Production and Systems
Waseda University
2-7 Hibikino, Wakamatsu-ku,
Kitakyushu-shi, Fukuoka-ken, 808-0135, Japan
Tel/Fax: +81-93-692-5287