| Sumario: | The field of distributional semantics has seen significant progress in recent years due to advancements in natural language processing techniques, particularly through the development of Neural Language Models like GPT and BERT. However, there are still challenges to overcome in terms of semantic representation, particularly in the lack of coherence and consistency in existing representation systems. This work introduces a framework defining the relationship between a probabilistic space, a set of meanings, and a vector space of static embedding representations; and establishes formal properties based on definitions that would be desirable for any distributional representation system to comply with in order to establish a common ground between distributional semantics and other approaches. This work also introduces an evaluation benchmark, defined on the basis of the formal properties introduced, which will allow to measure the quality of a representation system.
|
|---|
