Assessing Neural Referential Form Selectors on a Realistic Multilingual Dataset
This work addresses the problem of dataset limitations for researchers in natural language processing, particularly for multilingual referring expression generation, though it is incremental as it builds on existing REG/RFS methods with new data.
The authors tackled the limited scope of existing Neural Referring Expression Generation (REG) datasets by creating a multilingual dataset from OntoNotes, covering English and Chinese with a broader range of referring expression use. They built and assessed neural Referential Form Selection (RFS) models, finding that OntoNotes is better for evaluation than WebNLG and confirming that Chinese RFS relies more on discourse context than English, aligning with linguistic theories.
Previous work on Neural Referring Expression Generation (REG) all uses WebNLG, an English dataset that has been shown to reflect a very limited range of referring expression (RE) use. To tackle this issue, we build a dataset based on the OntoNotes corpus that contains a broader range of RE use in both English and Chinese (a language that uses zero pronouns). We build neural Referential Form Selection (RFS) models accordingly, assess them on the dataset and conduct probing experiments. The experiments suggest that, compared to WebNLG, OntoNotes is better for assessing REG/RFS models. We compare English and Chinese RFS and confirm that, in line with linguistic theories, Chinese RFS depends more on discourse context than English.