On the learnability of aspectual usage

We examine the learnability of grammatical aspect in Polish through the application of an error-driven learning algorithm to corpus data. We explore whether mastery of aspectual usage can be attained solely through exposure to co-occurrence patterns, or if an understanding of the abstract semantic distinctions discussed in aspectual literature are a prerequisite. To validate the model, we compare the corpus-based findings with data on aspectual usage from a survey of Polish L1 users. Our findings show that aspectual preferences are best predicted by a model that relies on co-occurrence information to predict a particular lemma, without reference to aspectual information. The models incorporating aspectual information, either by predicting aspect from co-occurrence information or from abstract semantic labels performed adequately but were only able to capture usage preferences relating to one of the two aspects. This study contributes insights into the role of usage in the acquisition linguistic categories and underscores the importance of integrating diverse evidence sources and methodologies for the development and validation of linguistic theories.