What can Neural Referential Form Selectors Learn?
This work addresses transparency issues in neural Referring Expression Generation models, providing insights into feature learning for researchers in computational linguistics and natural language processing, though it is incremental as it focuses on probing existing models.
The study investigated the extent to which neural Referential Form Selection models learn linguistic features influencing referring expression form, finding that all defined features were learned to some extent, with highest performance for referential status and syntactic position and lowest for discourse structure properties beyond the sentence level.
Despite achieving encouraging results, neural Referring Expression Generation models are often thought to lack transparency. We probed neural Referential Form Selection (RFS) models to find out to what extent the linguistic features influencing the RE form are learnt and captured by state-of-the-art RFS models. The results of 8 probing tasks show that all the defined features were learnt to some extent. The probing tasks pertaining to referential status and syntactic position exhibited the highest performance. The lowest performance was achieved by the probing models designed to predict discourse structure properties beyond the sentence level.