Mapping the Timescale Organization of Neural Language Models
This research provides a novel method for understanding the functional organization of recurrent neural networks, particularly for researchers and developers working on natural language processing models, by revealing how contextual information is processed over varying timescales.
This paper investigates how recurrent neural networks (RNNs) process contextual information over different timescales, similar to the human brain's hierarchical language processing. By applying neuroscience tools to an LSTM language model, the authors identified a small subset of units (less than 15%) with long processing timescales, which they further categorized into 'controller' and 'integrator' units based on their connectivity and functional impact on model performance at different sentence positions.
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. Additionally, the mapping revealed a small subset of the network (less than 15% of units) with long timescales and whose function had not previously been explored. We next probed the functional organization of the network by examining the relationship between the processing timescale of units and their network connectivity. We identified two classes of long-timescale units: "controller" units composed a densely interconnected subnetwork and strongly projected to 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 positions within 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 and models with different architectures. In summary, we demonstrated a model-free technique for mapping the timescale organization in recurrent neural networks, and we applied this method to reveal the timescale and functional organization of neural language models.