What they do when in doubt: a study of inductive biases in seq2seq learners
This work addresses the limited understanding of how seq2seq models generalize, which is crucial for improving their reliability in natural language processing and other sequence-based applications.
The study investigated the inductive biases of seq2seq learners in ambiguous tasks, finding that LSTM-based learners can learn counting, addition, and multiplication from a single example, while different architectures show preferences for hierarchical vs. linear or compositional vs. memorization generalization.
Sequence-to-sequence (seq2seq) learners are widely used, but we still have only limited knowledge about what inductive biases shape the way they generalize. We address that by investigating how popular seq2seq learners generalize in tasks that have high ambiguity in the training data. We use SCAN and three new tasks to study learners' preferences for memorization, arithmetic, hierarchical, and compositional reasoning. Further, we connect to Solomonoff's theory of induction and propose to use description length as a principled and sensitive measure of inductive biases. In our experimental study, we find that LSTM-based learners can learn to perform counting, addition, and multiplication by a constant from a single training example. Furthermore, Transformer and LSTM-based learners show a bias toward the hierarchical induction over the linear one, while CNN-based learners prefer the opposite. On the SCAN dataset, we find that CNN-based, and, to a lesser degree, Transformer- and LSTM-based learners have a preference for compositional generalization over memorization. Finally, across all our experiments, description length proved to be a sensitive measure of inductive biases.