On Neurons Invariant to Sentence Structural Changes in Neural Machine Translation
This work addresses the problem of understanding syntactic representations in neural models for researchers in machine translation and computational linguistics, but it is incremental as it builds on existing analysis methods without major breakthroughs.
The study investigated how neural machine translation models represent sentence structure by analyzing neuron activations in a Transformer model for English-German translation, finding that activation similarities are largely explained by word choice and sentence length rather than structural differences, but interventions on activations could somewhat control syntactic output forms.
We present a methodology that explores how sentence structure is reflected in neural representations of machine translation systems. We demonstrate our model-agnostic approach with the Transformer English-German translation model. We analyze neuron-level correlation of activations between paraphrases while discussing the methodology challenges and the need for confound analysis to isolate the effects of shallow cues. We find that similarity between activation patterns can be mostly accounted for by similarity in word choice and sentence length. Following that, we manipulate neuron activations to control the syntactic form of the output. We show this intervention to be somewhat successful, indicating that deep models capture sentence-structure distinctions, despite finding no such indication at the neuron level. To conduct our experiments, we develop a semi-automatic method to generate meaning-preserving minimal pair paraphrases (active-passive voice and adverbial clause-noun phrase) and compile a corpus of such pairs.