Iterative Document-level Information Extraction via Imitation Learning (2022-10-12T00:00:00.000000Z)

Abstract

We present a novel iterative extraction model, IterX, for extracting complex relations, or templates, i.e., N-tuples representing a mapping from named slots to spans of text within a document. Documents may feature zero or more instances of a template of any given type, and the task of template extraction entails identifying the templates in a document and extracting each template’s slot values. Our imitation learning approach casts the problem as a Markov decision process (MDP), and relieves the need to use predefined template orders to train an extractor. It leads to state-of-the-art results on two established benchmarks – 4-ary relation extraction on SciREX and template extraction on MUC-4 – as well as a strong baseline on the new BETTER Granular task.

Authors

Benjamin Van Durme

12 papers

Yunmo Chen

2 papers

Aaron Steven White

2 papers

Abstract

Authors

References59 items

On Event Individuation for Document-Level Information Extraction

An Empirical Study on Finding Spans

Everything Is All It Takes: A Multipronged Strategy for Zero-Shot Cross-Lingual Information Extraction

Document-level Entity-based Extraction as Template Generation

Template Filling with Generative Transformers

Document-Level Event Argument Extraction by Conditional Generation

LOME: Large Ontology Multilingual Extraction

Joint Modeling of Arguments for Event Understanding

A Joint Neural Model for Information Extraction with Global Features

SciREX: A Challenge Dataset for Document-Level Information Extraction

PyTorch: An Imperative Style, High-Performance Deep Learning Library

Multi-Sentence Argument Linking

Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer

DocRED: A Large-Scale Document-Level Relation Extraction Dataset

Improving Long Distance Slot Carryover in Spoken Dialogue Systems

Levenshtein Transformer

AMR Beyond the Sentence: the Multi-sentence AMR corpus

A Multi-sentiment-resource Enhanced Attention Network for Sentiment Classification

AllenNLP: A Deep Semantic Natural Language Processing Platform

End-to-end Neural Coreference Resolution

Attention is All you Need

Cross-Sentence N-ary Relation Extraction with Graph LSTMs

Bidirectional Attention Flow for Machine Comprehension

Distant Supervision for Relation Extraction beyond the Sentence Boundary

Scheduled Sampling for Sequence Prediction with Recurrent Neural Networks

Learning to Search Better than Your Teacher

Improving Multi-Step Prediction of Learned Time Series Models

On the Properties of Neural Machine Translation: Encoder–Decoder Approaches

Predicate-specific Annotations for Implicit Role Binding: Corpus Annotation, Data Analysis and Evaluation Experiments

A Dynamic Oracle for Arc-Eager Dependency Parsing

Modeling Textual Cohesion for Event Extraction

Peeling Back the Layers: Detecting Event Role Fillers in Secondary Contexts

A Reduction of Imitation Learning and Structured Prediction to No-Regret Online Learning

Beyond NomBank: A Study of Implicit Arguments for Nominal Predicates

A Unified Model of Phrasal and Sentential Evidence for Information Extraction

Search-based structured prediction

On Coreference Resolution Performance Metrics

Automatic Labeling of Semantic Roles

Message Understanding Conference- 6: A Brief History

MUC-4 evaluation metrics

Overview of the Third Message Understanding Evaluation and Conference

Algorithms for the Assignment and Transportation Problems

The Hungarian method for the assignment problem

GRIT: Generative Role-filler Transformers for Document-level Event Entity Extraction

D.3 BETTER Granular Evaluation for the BETTER Granular task bears some core similarity to CEAF-REEdef

2021), we frame this as a template extraction task, treating each 4-tuple as a template with four slots. D Model Evaluation Details A key consideration

2021a), has since been adopted as the standard evaluation metric for MUC-4

22This is precisely the hierarchy described for the informative argument extraction task

. GRIT: Generative role-ﬁller transformers

BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding

ACE 2005 Multilingual Training Corpus LDC2006T06

x ∈ S such that x ∈ R and that φ(R, S) = 1 when R = S, for reasons described in Luo

Empirical Methods in Natural Language Processing

Semeval-2010 Task 10: Linking Events and Their Participants in Discourse

some hazard in introducing generalized extraction (GTE) into this landscape.

Figure 1 and Figure 2. Template anchors are annotated either for MUC-4 or for S CI REX. they are annotated for BETTER Granular not factor into scoring

EE requires identifying lexical event triggers, whereas GTE does not, as template instances do not necessarily have one speciﬁc lexical anchor

document describing an epidemic may only explicitly state that thousands of people have contracted a particular disease

typed lexical triggers must be extracted. Second, we take GTE to be a fundamentally document-level task: templates concern events described over an entire document

Field of Study

Venue Information

Name

Type

URL

Alternate Names

x ∈ S such that x ∈ R and that φ(R, S) = 1 when R = S, for reasons described in Luo