On many occasions we are faced with problems whose answer is not easy to locate. These questions may be related to different domains, such as the medical domain, the legal domain or others. In these cases, tools that allow us to locate the concepts of each domain and the relationships between them are of great help both for decision makers in each domain, as well as for non-expert consumers.
The extraction of relationships between the main concepts of each domain is one of the tasks within natural language processing (NLP) that facilitate the generation of this knowledge. However, this topic presents numerous challenges such as the low performances of models in domain-specific texts, the limited availability of annotated corpora or their small size, and the dominant presence of the English language in comparison with Spanish language models and corpora. For this reason, the present work aims at improving relation extraction models for small domain-specific corpora.
Additionally, the legal domain was selected as a case study for the development of the project due to the relevance of the topic for general audiences as well as for information retrieval and question-answering tasks. Such applications are highly demanded by the population.