Severing the Edge Between Before and After: Neural Architectures for Temporal Ordering of Events

Abstract

In this paper, we propose a neural architecture and a set of training methods for ordering events by predicting temporal relations. Our proposed models receive a pair of events within a span of text as input and they identify temporal relations (Before, After, Equal, Vague) between them. Given that a key challenge with this task is the scarcity of annotated data, our models rely on either pretrained representations (i.e. RoBERTa, BERT or ELMo), transfer and multi-task learning (by leveraging complementary datasets), and self-training techniques. Experiments on the MATRES dataset of English documents establish a new state-of-the-art on this task.

TIMERS: Document-level Temporal Relation Extraction

“…Table 3 compares our work to the baseline methods reported on the TDDMan, TDDAuto, MATRES, and TimeBank-Dense datasets. We also include results for BERT-based Transformer (Devlin et al, 2019) and RoBERTa (Liu et al, 2019) following Ballesteros et al (2020). To prevent truncation or memory errors otherwise caused by multi-sentence spans, we concatenate only sentences containing source and events as input to Transformer baselines.…”

“…Prior work focuses on extracting temporal relations between event pairs (a.k.a., TLINKS) present in the same sentence (Intra-sentence TLINKS) or adjacent sentences (Inter-sentence TLINKS), mostly ignoring document-level pairs (Crossdocument TLINKS) (Reimers et al, 2016). Past works have used RNN (Cheng and Miyao, 2017;Meng et al, 2017;Goyal and Durrett, 2019;Ning et al, 2019;Han et al, 2019aHan et al, ,c,b, 2020b and Transformer networks (Ballesteros et al, 2020;Zhao et al, 2020b) for encoding a few sentences or a short paragraph but do not capture longrange dependencies and multi-hop reasoning at the document-level. This shortcoming is shown in the TDDiscourse dataset (Naik et al, 2019), which was designed to highlight global discourse-level challenges, e.g., multi-hop chain reasoning, future or hypothetical events, and reasoning requiring world knowledge.…”

“…Results comparing performance of TIMERS with baselines and ablative components on TDDMan, TDDAuto, MATRES and TimeBank-Dense datasets. We adopt the BERT and RoBERTa implementation from(Ballesteros et al, 2020). * indicates statistical significance over BERT Transformer (p ≤ 0.005) under Wilcoxon's Signed Rank test.…”

Neural Language Modeling for Contextualized Temporal Graph Generation

“…We note that the problem of temporal graph extraction is different from the more popular task of Temporal relation extraction (Temprel), which deals with classifying the temporal link between two already extracted events. State of the art Temprel systems use neural methods (Ballesteros et al, 2020;Ning et al, 2019b;Goyal and Durrett, 2019;Han et al, 2019;Cheng and Miyao, 2017), but typically use a handful of documents for their development and evaluation. Vashishtha et al (2019) are a notable exception by using Amazon Mechanical Turks to obtain manual annotations over a larger dataset of 16,000 sentences.…”

CoCoLM: COmplex COmmonsense Enhanced Language Model with Discourse Relations

“…The tasks are binary or multiple classification problems. Note the dataset of MATRES is split at the article level as in the previous work[15].2. MATRES[18] is a pairwise event temporal ordering prediction dataset, where each event pair in one document is annotated with a temporal relation (Before, After, Equal, Vague).…”