Probing Toxic Content in Large Pre-Trained Language Models (2021-01-01T00:00:00.000000Z)

TL;DR

A method based on logistic regression classifiers is proposed to probe English, French, and Arabic PTLMs and quantify the potentially harmful content that they convey with respect to a set of templates to assess and mitigate the toxicity transmitted by PTL Ms.

Abstract

Large pre-trained language models (PTLMs) have been shown to carry biases towards different social groups which leads to the reproduction of stereotypical and toxic content by major NLP systems. We propose a method based on logistic regression classifiers to probe English, French, and Arabic PTLMs and quantify the potentially harmful content that they convey with respect to a set of templates. The templates are prompted by a name of a social group followed by a cause-effect relation. We use PTLMs to predict masked tokens at the end of a sentence in order to examine how likely they enable toxicity towards specific communities. We shed the light on how such negative content can be triggered within unrelated and benign contexts based on evidence from a large-scale study, then we explain how to take advantage of our methodology to assess and mitigate the toxicity transmitted by PTLMs.

Authors

Dit-Yan Yeung

3 papers

Nedjma Djouhra Ousidhoum

3 papers

Yangqiu Song

11 papers

TL;DR

Abstract

Authors

References58 items

HONEST: Measuring Hurtful Sentence Completion in Language Models

Do language models learn typicality judgments from text?

On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? 🦜

HateXplain: A Benchmark Dataset for Explainable Hate Speech Detection

Comparative Evaluation of Label Agnostic Selection Bias in Multilingual Hate Speech Datasets

Social Chemistry 101: Learning to Reason about Social and Moral Norms

Multilingual Offensive Language Identification with Cross-lingual Embeddings

RealToxicityPrompts: Evaluating Neural Toxic Degeneration in Language Models

Would you Rather? A New Benchmark for Learning Machine Alignment with Cultural Values and Social Preferences

Probing for Referential Information in Language Models

SemEval-2020 Task 12: Multilingual Offensive Language Identification in Social Media (OffensEval 2020)

Language (Technology) is Power: A Critical Survey of “Bias” in NLP

Beyond Accuracy: Behavioral Testing of NLP Models with CheckList

Gender Gap in Natural Language Processing Research: Disparities in Authorship and Citations

Birds Have Four Legs?! NumerSense: Probing Numerical Commonsense Knowledge of Pre-trained Language Models

Social Biases in NLP Models as Barriers for Persons with Disabilities

When BERT Plays the Lottery, All Tickets Are Winning

StereoSet: Measuring stereotypical bias in pretrained language models

Deep Learning Models for Multilingual Hate Speech Detection

A Primer in BERTology: What We Know About How BERT Works

AraBERT: Transformer-based Model for Arabic Language Understanding

Bias in word embeddings

CamemBERT: a Tasty French Language Model

Predictive Biases in Natural Language Processing Models: A Conceptual Framework and Overview

Unintended Bias in Misogyny Detection

The Woman Worked as a Babysitter: On Biases in Language Generation

Language Models as Knowledge Bases?

Multilingual and Multi-Aspect Hate Speech Analysis

What BERT Is Not: Lessons from a New Suite of Psycholinguistic Diagnostics for Language Models

RoBERTa: A Robustly Optimized BERT Pretraining Approach

Social Data: Biases, Methodological Pitfalls, and Ethical Boundaries

The Risk of Racial Bias in Hate Speech Detection

We Need to Talk about Standard Splits

Women’s Syntactic Resilience and Men’s Grammatical Luck: Gender-Bias in Part-of-Speech Tagging and Dependency Parsing

Mitigating Gender Bias in Natural Language Processing: Literature Review

Measuring Bias in Contextualized Word Representations

A Structural Probe for Finding Syntax in Word Representations

What do you learn from context? Probing for sentence structure in contextualized word representations

Racial Bias in Hate Speech and Abusive Language Detection Datasets

Probing Biomedical Embeddings from Language Models

SemEval-2019 Task 6: Identifying and Categorizing Offensive Language in Social Media (OffensEval)

Measuring and Mitigating Unintended Bias in Text Classification

Transformers

Targeted Syntactic Evaluation of Language Models

Reducing Gender Bias in Abusive Language Detection

Are they Our Brothers? Analysis and Detection of Religious Hate Speech in the Arabic Twittersphere

Ultra-Fine Entity Typing

Examining Gender and Race Bias in Two Hundred Sentiment Analysis Systems

Large Scale Crowdsourcing and Characterization of Twitter Abusive Behavior

Word embeddings quantify 100 years of gender and ethnic stereotypes

Men Also Like Shopping: Reducing Gender Bias Amplification using Corpus-level Constraints

Automated Hate Speech Detection and the Problem of Offensive Language

“Why Should I Trust You?”: Explaining the Predictions of Any Classifier

Language Models are Unsupervised Multitask Learners

ATOMIC: an atlas of machine commonsense for if-then reasoning

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

L-HSAB: A Levantine Twitter Dataset for Hate Speech and Abusive Language

, Hannah Rashkin , Brendan Roof , Noah A . Smith , and Yejin Choi . 2019 b . ATOMIC : an atlas of machine common - sense for if - then reasoning

Field of Study

Venue Information

Name

Type

URL

Alternate Names