We analyze the learning dynamics of neural language and translation models using Loss Change Allocation (LCA), an indicator that enables a fine-grained analysis of parameter updates when optimizing for the loss function. In other words, we can observe the contributions of different network components at training time. In this article, we systematically study masked language modeling, causal language modeling, and machine translation. We show that the choice of training objective leads to distinctive optimization procedures, even when performed on comparable Transformer architectures. We demonstrate how the various Transformer parameters are used during training, supporting that the feed-forward components of each layer are the main contributors to the optimization procedure. Finally, we find that the learning dynamics are not affected by data size and distribution but rather determined by the learning objective.

Various studies show that pretrained language models such as BERT cannot straightforwardly replace encoders in neural machine translation despite their enormous success in other tasks. This is even more astonishing considering the similarities between the architectures. This paper sheds some light on the embedding spaces they create, using average cosine similarity, contextuality metrics and measures for representational similarity for comparison, revealing that BERT and NMT encoder representations look significantly different from one another. In order to address this issue, we propose a supervised transformation from one into the other using explicit alignment and fine-tuning. Our results demonstrate the need for such a transformation to improve the applicability of BERT in MT.

Contextualized word representations encode rich information about syntax and semantics, alongside specificities of each context of use. While contextual variation does not always reflect actual meaning shifts, it can still reduce the similarity of embeddings for word instances having the same meaning. We explore the imprint of two specific linguistic alternations, namely passivization and negation, on the representations generated by neural models trained with two different objectives: masked language modeling and translation. Our exploration methodology is inspired by an approach previously proposed for removing societal biases from word vectors. We show that passivization and negation leave their traces on the representations, and that neutralizing this information leads to more similar embeddings for words that should preserve their meaning in the transformation. We also find clear differences in how the respective features generalize across datasets.

In this paper we introduce a new natural language processing dataset and benchmark for predicting prosodic prominence from written text. To our knowledge this will be the largest publicly available dataset with prosodic labels. We describe the dataset construction and the resulting benchmark dataset in detail and train a number of different models ranging from feature-based classifiers to neural network systems for the prediction of discretized prosodic prominence. We show that pre-trained contextualized word representations from BERT outperform the other models even with less than 10% of the training data. Finally we discuss the dataset in light of the results and point to future research and plans for further improving both the dataset and methods of predicting prosodic prominence from text. The dataset and the code for the models will be made publicly available.