Mind the Gap: Assessing Temporal Generalization in Neural Language Models

Our world is open-ended, non-stationary, and constantly evolving; thus what we talk about and how we talk about it change over time. This inherent dynamic nature of language contrasts with the current static language modelling paradigm, which trains and evaluates models on utterances from overlapping time periods. Despite impressive recent progress, we demonstrate that Transformer-XL language models perform worse in the realistic setup of predicting future utterances from beyond their training period, and that model performance becomes increasingly worse with time. We find that, while increasing model size alone-a key driver behind recent progress-does not solve this problem, having models that continually update their knowledge with new information can indeed mitigate this performance degradation over time. Hence, given the compilation of ever-larger language modelling datasets, combined with the growing list of language-model-based NLP applications that require up-to-date factual knowledge about the world, we argue that now is the right time to rethink the static way in which we currently train and evaluate our language models, and develop adaptive language models that can remain up-to-date with respect to our ever-changing and non-stationary world. We will publicly release our dynamic, streaming language modelling benchmarks for WMT and ARXIV to facilitate language model evaluation that takes temporal dynamics into account. 1 * Equal contribution. ♠ Project initiation. Paper writing. ♦ Project technical infrastructure. ♥ Model design and experiments. ♣ Project support and advice. 1 We release our dynamic (streaming) language modelling benchmark for WMT and ARXIV at https: //github.com/deepmind/deepmind-research/tree/master/pitfalls_static_language_models . 2 In the case of GPT-3 (Brown et al., 2020), such tasks include LAMBADA (Paperno et al., 2016 ), TriviaQA (Joshi et al., 2017b), and WMT translation datasets, among others. These tasks were introduced between 2014 and 2017, which overlap in time with the GPT-3 CommonCrawl dataset that covered the period of 2016-2019.