Mapping the Timescale Organization of Neural Language Models
In the human brain, sequences of language input are processed within a distributed and hierarchical architecture, in which higher stages of processing encode contextual information over longer timescales. In contrast, in recurrent neural networks which perform natural language processing, we know little about how the multiple timescales of contextual information are functionally organized. Therefore, we applied tools developed in neuroscience to map the "processing timescales" of individual units within a word-level LSTM language model. This timescale-mapping method assigned long timescales to units previously found to track long-range syntactic dependencies, and revealed a new cluster of previously unreported long-timescale units. Next, we explored the functional role of units by examining the relationship between their processing timescales and network connectivity. We identified two classes of long-timescale units: "Controller" units composed a densely interconnected subnetwork and strongly projected to the forget and input gates of the rest of the network, while "Integrator" units showed the longest timescales in the network, and expressed projection profiles closer to the mean projection profile. Ablating integrator and controller units affected model performance at different position of a sentence, suggesting distinctive functions of these two sets of units. Finally, we tested the generalization of these results to a character-level LSTM model. In summary, we demonstrated a model-free technique for mapping the timescale organization in neural network models, and we applied this method to reveal the timescale and functional organization of LSTM language models.
READ FULL TEXT