Unsupervised inference of implicit biomedical events using context triggers

Abstract

Background: Event extraction from the biomedical literature is one of the most actively researched areas in biomedical text mining and natural language processing. However, most approaches have focused on events within single sentence boundaries, and have thus paid much less attention to events spanning multiple sentences. The Bacteria-Biotope event (BB-event) subtask presented in BioNLP Shared Task 2016 is one such example; a significant amount of relations between bacteria and biotope span more than one sentence, but existing systems have treated them as false negatives because labeled data is not sufficiently large enough to model a complex reasoning process using supervised learning frameworks. Results: We present an unsupervised method for inferring cross-sentence events by propagating intra-sentence information to adjacent sentences using context trigger expressions that strongly signal the implicit presence of entities of interest. Such expressions can be collected from a large amount of unlabeled plain text based on simple syntactic constraints, helping to overcome the limitation of relying only on a small number of training examples available. The experimental results demonstrate that our unsupervised system extracts cross-sentence events quite well and outperforms all the state-of-the-art supervised systems when combined with existing methods for intra-sentence event extraction. Moreover, our system is also found effective at detecting long-distance intra-sentence events, compared favorably with existing high-dimensional models such as deep neural networks, without any supervised learning techniques. Conclusions: Our linguistically motivated inference model is shown to be effective at detecting implicit events that have not been covered by previous work, without relying on training data or curated knowledge bases. Moreover, it also helps to boost the performance of existing systems by allowing them to detect additional cross-sentence events. We believe that the proposed model offers an effective way to infer implicit information beyond sentence boundaries, especially when human-annotated data is not sufficient enough to train a robust supervised system.

A Survey on Event Extraction for Natural Language Understanding: Riding the Biomedical Literature Wave

“…An event may involve arguments that are distant from each other or from the trigger, peculiarly frequent in long documents such as biomedical publications. Compared to RE, an event extent might be composed of many sentences, and it is hard to find all arguments of one event instance in a single sentence (arguments-scattering) [73]. Simply put, events do not necessarily constitute continuous spans of text, and their components are fragmented.…”

“…Closed-domain [69]- [72], [85], [89], [100], [115], [125], [133], [134], [136]- [138], [142], [143], [145], [148], [159]- [161], [165]- [167], [169], [180]- [185], [201], [202], [216]- [220], [229]- [233], [241] [186] [42]- [44], [46], [47], [58], [60], [65], [73], [99], [114], [135], [139], [140], [149], [162], [163], [177]- [179], [187]- [191], [199], [203]- [205...…”

“…Bui et al [177] Pattern learning \w structured mapping Pham et al [187] Hybrid (Pattern matching + ML) Liu et al [188] Hybrid (Pattern matching + ML) Venugopal et al [190] Hybrid (MLN + SVM) Björne and Salakoski [237] SVM Liu et al [189] Hybrid (graph-based patterns + ML) [205] Bi-GRU Li et al [99] Tree-LSTM Li et al [60] Multi-pooling CNN Zhu and Zheng [47] Hybrid (CNN + RNN) Trieu et al [44] SciBERT + custom layers Chung et al [73] Patterns \w context triggers Wang et al [135] EE as QA Huang et al [139] SciBERT + GNN Ramponi et al [140] EE as sequence labeling Yu et al [204] Bi-LSTM + Tree-LSTM Zhang et al [186] Hybrid (Bi-LSTM + GCN)…”

“…Most research has merely focused on extracting events within the sentence scope, where a single sentence is assumed to cover complete pieces of information. Per contra, writers usually adopt several even non-continuous sentences to articulate individual or multiple events; this also applies to many BioNLP-ST datasets [73]. In this sense, much less attention has been paid to interactions discussed across a whole docu-…”

“…Alternatively, Gupta et al [286] used recurrent and recursive neural networks to classify the type of simple binary biomedical events within and across sentence boundaries, leveraging on inter-sentential dependency paths between entity pairs. Chung et al [73] noted that if an entity is mentioned in a sentence, it is possible to find expressions in its adjacent sentences implying the presence of that entity. They called these expressions "context triggers" (example in Table 8).…”