Text2Event: Controllable Sequence-to-Structure Generation for End-to-end Event Extraction

By Yaojie Lu, Hongyu Lin, Jin Xu, Xianpei Han, Jialong Tang, Annan Li, Le Sun, Meng Liao, Shaoyi Chen. Annual Meeting of the Association for Computational Linguistics, 2021

338 citations recorded by the source corpus.

Event extraction is challenging due to the complex structure of event records and the semantic gap between text and event. Traditional methods usually extract event records by decomposing the complex structure prediction task into multiple subtasks. In this paper, we propose Text2Event, a sequence-to-structure generation paradigm that can directly extract events from the text in an end-to-end manner. Specifically, we design a sequence-to-structure network for unified event extraction, a constrained decoding algorithm for event knowledge injection during inference, and a curriculum learning algorithm for efficient model learning. Experimental results show that, by uniformly modeling all tasks in a single model and universally predicting different labels, our method can achieve competitive performance using only record-level annotations in both supervised learning and transfer learning settings.

Subfields: Artificial Intelligence, Molecular Biology

View Paper: https://www.semanticscholar.org/paper/a0c78661e4a5278c65cefc5b4fdef669dcae7e17

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Lifu Huang
2. Shiyu Chang
3. Mo Yu
4. Sijia Wang
5. Heng Ji
6. Xingyao Wang
7. Sha Li
8. Xianpei Han

In Molecular Biology

1. Heng Ji
2. Mo Yu
3. Shiyu Chang
4. Lichao Sun

Citations in the corpus

• Code4Struct: Code Generation for Few-Shot Structured Prediction from Natural Language (2022, background): “There is a tradeoff between the effort put into defining the output or designing the prompt ( e.g. , Text2Event in Table 2) and the benefit from pre-training in natural language ( e.g. , DEGREE and BART-Gen in Table 2).”
• Large language models for generative information extraction: a survey (2023, neutral): “Additionally, a notable trend is the emergence of unified models for IE tasks, leveraging the general text understanding capabilities of modern large models [4, 6, 156].”
• Zero-Shot Information Extraction via Chatting with ChatGPT (2023, neutral): “The baselines are PaddleNLP LIC2021 IE 5 and CaseRel (Wei et al., 2020) (Lu et al., 2021) for EE.”
• Unified Structure Generation for Universal Information Extraction (2022, methodology): “We employ the state-of-the-art generation-based event extraction method T EXT 2E VENT (Lu et al., 2021) as the comparable state-of-the-art system.”
• Summary of ChatGPT-Related Research and Perspective Towards the Future of Large Language Models (2023, methodology): “Gao et al. [19] explored the feasibility and challenges of ChatGPT for event extraction on the ACE2005 corpus, evaluating the performance of ChatGPT in long-tail and complex scenarios (texts containing multiple events) and comparing it with two task-specific models, Text2Event and EEQA [57; 14].”
• Query and Extract: Refining Event Extraction as Type-oriented Binary Decoding (2021, methodology): “The statistics for the ERE 825 dataset is slightly different from previous work (Lin 826 et al., 2020; Lu et al., 2021) as we consider the 827 event triggers that are annotated with multiple types 828 as different instances while the previous studies829 just keep one annotated type for each trigger…”
• Summary of ChatGPT/GPT-4 Research and Perspective Towards the Future of Large Language Models (2023, methodology): “The results show that the average performance of Chat-GPT in long-tail and complex scenarios is only 51.04% of that of task-specific models such as EEQA.”