Classification and Clustering of Arguments with Contextualized Word Embeddings

@article{Reimers2019ClassificationAC,
  title={Classification and Clustering of Arguments with Contextualized Word Embeddings},
  author={Nils Reimers and Benjamin Schiller and Tilman Beck and Johannes Daxenberger and Christian Stab and Iryna Gurevych},
  journal={ArXiv},
  year={2019},
  volume={abs/1906.09821}
}
We experiment with two recent contextualized word embedding methods (ELMo and BERT) in the context of open-domain argument search. For the first time, we show how to leverage the power of contextualized word embeddings to classify and cluster topic-dependent arguments, achieving impressive results on both tasks and across multiple datasets. For argument classification, we improve the state-of-the-art for the UKP Sentential Argument Mining Corpus by 20.8 percentage points and for the IBM Debater… 

Tables from this paper

Fine-Grained Argument Unit Recognition and Classification
TLDR
This work presents a dataset of arguments from heterogeneous sources annotated as spans of tokens within a sentence, as well as with a corresponding stance, and shows that and how such difficult argument annotations can be effectively collected through crowdsourcing with high inter-annotator agreement.
An Empirical Study on Measuring the Similarity of Sentential Arguments with Language Model Domain Adaptation
TLDR
This work adapted a pretrained language model to a domain of interest using self-supervised learning, and fine-tuned the model for a task of measuring the similarity between sentences taken from different domains, suggesting the possibility of a generalized argument clustering model for various argumentative topics.
Aspect-Based Argument Mining
TLDR
This work creates and releases an annotated corpus with aspect information on the token-level, and considers aspects as the main point(s) argument units are addressing, for further downstream tasks such as argument ranking, argument summarization and generation, as well as the search for counter-arguments on the aspect-level.
Enhancing Legal Argument Mining with Domain Pre-training and Neural Networks
TLDR
A broad study of both classic and contextual embedding models and their performance on practical case law from the European Court of Human Rights (ECHR) is provided and a number of neural networks when being combined with different embeddings are explored.
A Large-scale Dataset for Argument Quality Ranking: Construction and Analysis
TLDR
This work addresses the core issue of inducing a labeled score from crowd annotations by performing a comprehensive evaluation of different approaches to this problem, and presents a neural method for argument quality ranking, which outperforms several baselines on the authors' own dataset, as well as previous methods published for another dataset.
Fine and Coarse Granular Argument Classification before Clustering
TLDR
A novel pipeline for clustering arguments is introduced and a new dataset consisting of 365 argument graphs accompanying more than 11,000 high-quality arguments that, contrary to previous datasets, have been generated, displayed, and verified by journalists and were published in newspapers are introduced.
DirectProbe: Studying Representations without Classifiers
TLDR
A heuristic is developed that directly studies the geometry of a representation by building upon the notion of a version space for a task, and can shine lights on how an embedding space represents labels and also anticipate the classifier performance for the representation.
Self-trained Pretrained Language Models for Evidence Detection
TLDR
Experiments on sentence level evidence detection show that self-training can complement pretrained language models to provide performance improvements, and a thorough investigation on how to utilize pseudo labels effectively in the self- training scheme is provided.
Transfer Learning with Sentence Embeddings for Argumentative Evidence Classication
TLDR
Results show that, even with small amount of data, classifiers trained on sentence embeddings extracted from pre-trained transformers can achieve encouraging scores, outperforming previous results on evidence classification.
Automatic Argument Quality Assessment - New Datasets and Methods
TLDR
This work explores the task of automatic assessment of argument quality, and suggests neural methods based on a recently released language model for argument ranking as well as for argument-pair classification that significantly outperform earlier methods.
...
...

References

SHOWING 1-10 OF 24 REFERENCES
Deep Contextualized Word Representations
TLDR
A new type of deep contextualized word representation is introduced that models both complex characteristics of word use and how these uses vary across linguistic contexts, allowing downstream models to mix different types of semi-supervision signals.
Cross-topic Argument Mining from Heterogeneous Sources
TLDR
This paper proposes a new sentential annotation scheme that is reliably applicable by crowd workers to arbitrary Web texts and shows that integrating topic information into bidirectional long short-term memory networks outperforms vanilla BiLSTMs in F1 in two- and three-label cross-topic settings.
Retrieval of the Best Counterargument without Prior Topic Knowledge
TLDR
This work hypothesizes the best counterargument to invoke the same aspects as the argument while having the opposite stance, and simultaneously model the similarity and dissimilarity of pairs of arguments, based on the words and embeddings of the arguments’ premises and conclusions.
GloVe: Global Vectors for Word Representation
TLDR
A new global logbilinear regression model that combines the advantages of the two major model families in the literature: global matrix factorization and local context window methods and produces a vector space with meaningful substructure.
Measuring the Similarity of Sentential Arguments in Dialogue
TLDR
It is shown here that the goal of this work is to induce and identify argument facets across multiple conversations, and produce summaries, and it is shown that it can predict ARGUMENT FACET SIMILARITY with a correlation averaging 0.63 compared to a human topline.
Neural End-to-End Learning for Computational Argumentation Mining
TLDR
This work investigates neural techniques for end-to-end computational argumentation mining and finds that jointly learning 'natural' subtasks, in a multi-task learning setup, improves performance.
Context Dependent Claim Detection
TLDR
This work formally defines the challenging task of automatic claim detection in a given context and outlines a preliminary solution, and assess its performance over annotated real world data, collected specifically for that purpose over hundreds of Wikipedia articles.
Supervised Learning of Universal Sentence Representations from Natural Language Inference Data
TLDR
It is shown how universal sentence representations trained using the supervised data of the Stanford Natural Language Inference datasets can consistently outperform unsupervised methods like SkipThought vectors on a wide range of transfer tasks.
Identifying Argumentative Discourse Structures in Persuasive Essays
TLDR
A novel approach for identifying argumentative discourse structures in persuasive essays by evaluating several classifiers and proposing novel feature sets including structural, lexical, syntactic and contextual features.
Identifying Prominent Arguments in Online Debates Using Semantic Textual Similarity
TLDR
A cluster analysis using semantic textual similarity to detect similar arguments in online debates and a preliminary cluster evaluation and error analysis based on cluster-class matching against a manually labeled dataset is performed.
...
...