Textual Entailment for Event Argument Extraction: Zero- and Few-Shot with Multi-Source Learning

@inproceedings{Sainz2022TextualEF,
  title={Textual Entailment for Event Argument Extraction: Zero- and Few-Shot with Multi-Source Learning},
  author={Oscar Sainz and Itziar Gonzalez-Dios and Oier Lopez de Lacalle and Bonan Min and Eneko Agirre},
  booktitle={NAACL-HLT},
  year={2022}
}
Recent work has shown that NLP tasks such as Relation Extraction (RE) can be recasted as Textual Entailment tasks using verbalizations, with strong performance in zero-shot and few-shot settings thanks to pre-trained entailment models. The fact that relations in current RE datasets are easily verbalized casts doubts on whether entailment would be effective in more complex tasks. In this work we show that entailment is also effective in Event Argument Extraction (EAE), reducing the need of… 

Few-shot Information Extraction is Here: Pre-train, Prompt and Entail

This talk will present an approach that combines prompts and PLMs fine-tuned for textual entailment that yields state-of-the-art results on Information Extraction (IE) using only a small fraction of the annotations.

ZS4IE: A toolkit for Zero-Shot Information Extraction with simple Verbalizations

This demonstration introduces a new workflow where the analyst directly verbalizes the entities/relations, which are then used by a Textual Entailment model to perform zero-shot IE.

Stretching Sentence-pair NLI Models to Reason over Long Documents and Clusters

This work further explores the direct zero-shot applicability of NLI models to real applications, beyond the sentence-pair setting they were trained on, and develops new aggregation methods to allow operating over full documents, reaching state-of-the-art performance on the ContractNLI dataset.

Schema-aware Reference as Prompt Improves Data-Efficient Relational Triple and Event Extraction

A novel approach of schema-aware R eference A s P rompt ( RAP), which dynamically leverage schema and knowledge inherited from global (few-shot) training data for each sample, and employs a dynamic reference integration module to retrieve pertinent knowledge from the datastore as prompts during training and inference.

OpenStance: Real-world Zero-shot Stance Detection

The key challenge of OpenStance lies in the open-domain gen-eralization : learning a system with fully un-specific supervision but capable of generaliz-ing to any dataset, which is proposed to combine indirect supervision from textual entailment datasets, and weak supervision from data generated automatically by pre-trained Language Models.

Yes, No or IDK: The Challenge of Unanswerable Yes/No Questions

The Yes/No QA task is extended, adding questions with an IDK answer, and the contribution of training on other Natural Language Understanding tasks is studied, focusing on Extractive QA and Recognizing Textual Entailments.

Learning to Select from Multiple Options

This work first proposes a contextualized TE model (Context-TE) by appending other k options as the context of the current (P, H) modeling, which is able to learn more reliable decision for the H since it considers various context and speed up Context-TE by coming up with Parallel-TE, which learns the decisions of multiple options simultaneously.

References

SHOWING 1-10 OF 72 REFERENCES

Label Verbalization and Entailment for Effective Zero and Few-Shot Relation Extraction

This work reformulates relation extraction as an entailment task, with simple, hand-made, verbalizations of relations produced in less than 15 min per relation, showing that the few-shot systems are specially effective when discriminating between relations, and that the performance difference in low data regimes comes mainly from identifying no-relation cases.

Universal Natural Language Processing with Limited Annotations: Try Few-shot Textual Entailment as a Start

This work demonstrates that this framework enables a pretrained entailment model to work well on new entailment domains in a few-shot setting, and shows its effectiveness as a unified solver for several downstream NLP tasks such as question answering and coreference resolution when the end-task annotations are limited.

Document-Level Event Argument Extraction by Conditional Generation

A document-level neural event argument extraction model is proposed by formulating the task as conditional generation following event templates by creating the first end-to-end zero-shot event extraction framework.

Zero-shot Event Extraction via Transfer Learning: Challenges and Insights

This work explores the possibility of zero-shot event extraction by formulating it as a set of Textual Entailment and/or Question Answering queries, exploiting pretrained TE/QA models for direct transfer.

Entity, Relation, and Event Extraction with Contextualized Span Representations

This work examines the capabilities of a unified, multi-task framework for three information extraction tasks: named entity recognition, relation extraction, and event extraction (called DyGIE++) and achieves state-of-the-art results across all tasks.

ZS4IE: A toolkit for Zero-Shot Information Extraction with simple Verbalizations

This demonstration introduces a new workflow where the analyst directly verbalizes the entities/relations, which are then used by a Textual Entailment model to perform zero-shot IE.

An Improved Baseline for Sentence-level Relation Extraction

This paper revisits two problems that affect the performance of existing RE models, namely entity representation and noisy or ill-defined labels, and presents an improved RE baseline, incorporated with entity representations with typed markers.

Probing and Fine-tuning Reading Comprehension Models for Few-shot Event Extraction

This work proposes a reading comprehension framework for event extraction, and forms event detection as a textual entailment prediction problem, and argues that the event extraction models so trained can generalize poorly across domains and text genres.

Few-Shot Text Generation with Natural Language Instructions

GenPET, a method for text generation that is based on pattern-exploiting training, a recent approach for combining textual instructions with supervised learning that only works for classification tasks, is introduced.

Entailment as Few-Shot Learner

A new approach is proposed, named as EFL, that can turn small LMs into better few-shot learners, and improves the various existing SOTA few-shots learning methods by 12%, and yields competitive few- shot performance with 500 times larger models, such as GPT-3.
...