What has LeBenchmark Learnt about French Syntax?
This addresses the question of how low-level acoustic models acquire higher-level linguistic knowledge, which is incremental for speech processing research.
The paper investigates whether LeBenchmark, a pretrained acoustic model trained on 7k hours of French speech, encodes syntactic information, finding that it does, with syntactic information most extractable from middle layers before a sharp decrease.
The paper reports on a series of experiments aiming at probing LeBenchmark, a pretrained acoustic model trained on 7k hours of spoken French, for syntactic information. Pretrained acoustic models are increasingly used for downstream speech tasks such as automatic speech recognition, speech translation, spoken language understanding or speech parsing. They are trained on very low level information (the raw speech signal), and do not have explicit lexical knowledge. Despite that, they obtained reasonable results on tasks that requires higher level linguistic knowledge. As a result, an emerging question is whether these models encode syntactic information. We probe each representation layer of LeBenchmark for syntax, using the Orféo treebank, and observe that it has learnt some syntactic information. Our results show that syntactic information is more easily extractable from the middle layers of the network, after which a very sharp decrease is observed.