CUE Vectors: Modular Training of Language Models Conditioned on Diverse Contextual Signals
share
We propose a framework to modularize the training of neural language models that use diverse forms of sentence-external context (including metadata) by eliminating the need to jointly train sentence-external and within-sentence encoders. Our approach, contextual universal embeddings (CUE), trains LMs on one set of context, such as date and author, and adapts to novel metadata types, such as article title, or previous sentence. The model consists of a pretrained neural sentence LM, a BERT-based context encoder, and a masked transformer decoder that estimates LM probabilities using sentence-internal and sentence-external information. When context or metadata are unavailable, our model learns to combine contextual and sentence-internal information using noisy oracle unigram embeddings as a proxy. Real contextual information can be introduced later and used to adapt a small number of parameters that map contextual data into the decoder's embedding space. We validate the CUE framework on a NYTimes text corpus with multiple metadata types, for which the LM perplexity can be lowered from 36.6 to 27.4 by conditioning on context. Bootstrapping a contextual LM with only a subset of the context/metadata during training retains 85% of the achievable gain. Training the model initially with proxy context retains 67 context. Furthermore, we can swap one type of pretrained sentence LM for another without retraining the context encoders, by only adapting the decoder model. Overall, we obtain a modular framework that allows incremental, scalable training of context-enhanced LMs.
READ FULL TEXT