Adaptive Input Representations for Neural Language Modeling
This work addresses efficiency and performance bottlenecks in language modeling for NLP practitioners, offering a significant speedup and improved benchmarks, though it is incremental as it builds on prior adaptive methods.
The paper tackles the problem of efficient neural language modeling by introducing adaptive input representations that extend adaptive softmax to variable-capacity inputs, achieving over twice the training speed of character input CNN with fewer parameters and setting new state-of-the-art perplexity scores of 18.7 on WikiText-103 and 23.02 on the Billion Word benchmark.
We introduce adaptive input representations for neural language modeling which extend the adaptive softmax of Grave et al. (2017) to input representations of variable capacity. There are several choices on how to factorize the input and output layers, and whether to model words, characters or sub-word units. We perform a systematic comparison of popular choices for a self-attentional architecture. Our experiments show that models equipped with adaptive embeddings are more than twice as fast to train than the popular character input CNN while having a lower number of parameters. On the WikiText-103 benchmark we achieve 18.7 perplexity, an improvement of 10.5 perplexity compared to the previously best published result and on the Billion Word benchmark, we achieve 23.02 perplexity.