Machine learning is recently used to detect hate speech and other forms of abusive language in online platforms. However, a notable weakness of machine learning models is their vulnerability to bias, which can impair their performance and fairness. One type is annotator bias caused by the subjective perception of the annotators. In this work, we investigate annotator bias using classification models trained on data from demographically distinct annotator groups. To do so, we sample balanced subsets of data that are labeled by demographically distinct annotators. We then train classifiers on these subsets, analyze their performances on similarly grouped test sets, and compare them statistically. Our findings show that the proposed approach successfully identifies bias and that demographic features, such as first language, age, and education, correlate with significant performance differences.

/pdf/identifying-and-measuring-annotator-bias-based-on-annotators-1gbxzox9gy.pdf

Identifying and Measuring Annotator Bias Based on Annotators’ Demographic Characteristics

One challenge that social media platforms are facing nowadays is hate speech. Hence, automatic hate speech detection has been increasingly researched in recent years - in particular with the rise of deep learning. A problem of these models is their vulnerability to undesirable bias in training data. We investigate the impact of political bias on hate speech classification by constructing three politically-biased data sets (left-wing, right-wing, politically neutral) and compare the performance of classifiers trained on them. We show that (1) political bias negatively impairs the performance of hate speech classifiers and (2) an explainable machine learning model can help to visualize such bias within the training data. The results show that political bias in training data has an impact on hate speech classification and can become a serious issue.

/pdf/impact-of-politically-biased-data-on-hate-speech-1z48sgyemc.pdf

Impact of politically biased data on hate speech classification

Annotated data are essential to train and benchmark NLP systems. The reliability of the annotation, i.e. low interannotator disagreement, is a key factor, especially when dealing with highly subjective phenomena occurring in human language. Hate speech (HS), in particular, is intrinsically nuanced and hard to fit in any fixed scale, therefore crisp classification schemes for its annotation often show their limits. We test three annotation schemes on a corpus of HS, in order to produce more reliable data. While rating scales and best-worst-scaling are more expensive strategies for annotation, our experimental results suggest that they are worth implementing in a HS detection perspective.1

/pdf/annotating-hate-speech-three-schemes-at-comparison-2i26j5ug2b.pdf

Annotating hate speech: Three schemes at comparison

We present results for Subtask A and C of SemEval 2019 Shared Task 6. In Subtask A, we experiment with an embedding representation of postings and use BERT to categorize postings. Our best result reaches the 10th place (out of 103). In Subtask C, we applied a two-vote classification approach with minority fallback, which is placed on the 19th rank (out of 65).

/pdf/ltl-ude-at-semeval-2019-task-6-bert-and-two-vote-zc4hy2hdh2.pdf

LTL-UDE at SemEval-2019 Task 6: BERT and Two-Vote Classification for Categorizing Offensiveness.

In this paper, we present a new approach for the evaluation, error analysis, and interpretation of supervised and unsupervised Paraphrase Identification (PI) systems. Our evaluation framework makes use of a PI corpus annotated with linguistic phenomena to provide a better understanding and interpretation of the performance of various PI systems. Our approach allows for a qualitative evaluation and comparison of the PI models using human interpretable categories. It does not require modification of the training objective of the systems and does not place additional burden on the developers. We replicate several popular supervised and unsupervised PI systems. Using our evaluation framework we show that: 1) Each system performs differently with respect to a set of linguistic phenomena and makes qualitatively different kinds of errors; 2) Some linguistic phenomena are more challenging than others across all systems.

/pdf/a-qualitative-evaluation-framework-for-paraphrase-4uwh8i54zz.pdf

A Qualitative Evaluation Framework for Paraphrase Identification.

Understanding hate speech remains a significant challenge for both creating reliable datasets and automated hate speech detection. We hypothesize that being part of the targeted group or personally agreeing with an assertion substantially effects hate speech perception. To test these hypotheses, we create FEMHATE – a dataset containing 400 assertions that target women. These assertion are judged by female and male subjects for (i) how hateful these assertions are and (ii) for whether they agree with the assertions. We find that women and men consistently evaluate extreme cases of hate speech. We also find a strong relationship between hate speech and agreement judgments, showing that a low agreement score is a prerequisite for hate speech. We show how this relationship can be used for automatic hate speech detection. Our best system based on agreement judgments outperforms a baseline SVM classifier (equipped with ngrams) by a wide margin.

/pdf/do-women-perceive-hate-differently-examining-the-3mbdxqteto.pdf

Do Women Perceive Hate Differently: Examining the Relationship Between Hate Speech, Gender, and Agreement Judgments.

Pairs of sentences, phrases, or other text pieces can hold semantic relations such as paraphrasing, textual entailment, contradiction, specificity, and semantic similarity. These relations are usually studied in isolation and no dataset exists where they can be compared empirically. Here we present a corpus annotated with these relations and the analysis of these results. The corpus contains 520 sentence pairs, annotated with these relations. We measure the annotation reliability of each individual relation and we examine their interactions and correlations. Among the unexpected results revealed by our analysis is that the traditionally considered direct relationship between paraphrasing and bi-directional entailment does not hold in our data.

/pdf/annotating-and-analyzing-the-interactions-between-meaning-f6h4ex5sxz.pdf

Annotating and analyzing the interactions between meaning relations

We present a corpus of political debates annotated with aspect-based sentiment and a corpus analysis. The source corpus consists of transcribed speeches taken from the two presidential debates of the 2016 US election. We annotate the corpus according to two different schemata and analyze their differences. We show that the choice schema has a strong impact on the result of aspect-based sentiment analysis. Furthermore, we provide a corpus that can be used as a gold-standard for automatic aspect-based sentiment annotation of political debates.

/pdf/corpus-of-aspect-based-sentiment-in-political-debates-3c8ia2cyr8.pdf

Corpus of Aspect-based Sentiment in Political Debates

This paper describes our submission (winning solution for Task A) to the Shared Task on Hateful Meme Detection at WOAH 2021. We build our system on top of a state-of-the-art system for binary hateful meme classification that already uses image tags such as race, gender, and web entities. We add further metadata such as emotions and experiment with data augmentation techniques, as hateful instances are underrepresented in the data set.

/pdf/vl-bert-detecting-protected-groups-in-hateful-multimodal-rfg56oao0i.pdf

VL-BERT+: Detecting Protected Groups in Hateful Multimodal Memes

/pdf/corpus-of-aspect-based-sentiment-in-political-debates-tgg8030g7x.pdf

Darina Gold

Papers

Do Women Perceive Hate Differently: Examining the Relationship Between Hate Speech, Gender, and Agreement Judgments.

Annotating and analyzing the interactions between meaning relations

Corpus of Aspect-based Sentiment in Political Debates

VL-BERT+: Detecting Protected Groups in Hateful Multimodal Memes

Corpus of Aspect-based Sentiment in Political Debates.