A Hierarchical Transformer for Unsupervised Parsing

The underlying structure of natural language is hierarchical; words combine into phrases, which in turn form clauses. An awareness of this hierarchical structure can aid machine learning models in performing many linguistic tasks. However, most such models just process text sequentially and there is no bias towards learning hierarchical structure encoded into their architecture. In this paper, we extend the recent transformer model (Vaswani et al., 2017) by enabling it to learn hierarchical representations. To achieve this, we adapt the ordering mechanism introduced in Shen et al., 2018, to the self-attention module of the transformer architecture. We train our new model on language modelling and then apply it to the task of unsupervised parsing. We achieve reasonable results on the freely available subset of the WSJ10 dataset with an F1-score of about 50%.

Biomedical relation extraction with pre-trained language representations and minimal task-specific architecture

This paper presents our participation in the AGAC Track from the 2019 BioNLP Open Shared Tasks. We provide a solution for Task 3, which aims to extract "gene - function change - disease" triples, where "gene" and "disease" are mentions of particular genes and diseases respectively and "function change" is one of four pre-defined relationship types. Our system extends BERT (Devlin et al., 2018), a state-of-the-art language model, which learns contextual language representations from a large unlabelled corpus and whose parameters can be fine-tuned to solve specific tasks with minimal additional architecture. We encode the pair of mentions and their textual context as two consecutive sequences in BERT, separated by a special symbol. We then use a single linear layer to classify their relationship into five classes (four pre-defined, as well as 'no relation'). Despite considerable class imbalance, our system significantly outperforms a random baseline while relying on an extremely simple setup with no specially engineered features.

Constructing large scale biomedical knowledge bases from scratch with rapid annotation of interpretable patterns

Knowledge base construction is crucial for summarising, understanding and inferring relationships between biomedical entities. However, for many practical applications such as drug discovery, the scarcity of relevant facts (e.g. gene X is therapeutic target for disease Y) severely limits a domain expert's ability to create a usable knowledge base, either directly or by training a relation extraction model. In this paper, we present a simple and effective method of extracting new facts with a pre-specified binary relationship type from the biomedical literature, without requiring any training data or hand-crafted rules. Our system discovers, ranks and presents the most salient patterns to domain experts in an interpretable form. By marking patterns as compatible with the desired relationship type, experts indirectly batch-annotate candidate pairs whose relationship is expressed with such patterns in the literature. Even with a complete absence of seed data, experts are able to discover thousands of high-quality pairs with the desired relationship within minutes. When a small number of relevant pairs do exist - even when their relationship is more general (e.g. gene X is biologically associated with disease Y) than the relationship of interest - our system leverages them in order to i) learn a better ranking of the patterns to be annotated or ii) generate weakly labelled pairs in a fully automated manner. We evaluate our method both intrinsically and via a downstream knowledge base completion task, and show that it is an effective way of constructing knowledge bases when few or no relevant facts are already available.