Papers by Svetlana Kiritchenko
How Does Stereotype Content Differ across Data Sources? (2024.starsem-1)
Copied to clipboard
| Challenge: | Existing studies of stereotypes using rating scales capture beliefs and opinions about different social groups. |
| Approach: | They compare stereotype-relevant measures of social group social status with traditional scales and a word-list generation task using free-text data. |
| Outcome: | The results compare with traditional surveys and a spontaneous word-list generation task. |
Extracting Age-Related Stereotypes from Social Media Texts (2022.lrec-1)
Copied to clipboard
| Challenge: | a method for extracting age-related stereotypes from Twitter data is under-studied in NLP . stereotyping on the basis of protected characteristics has been understudied . |
| Approach: | They propose a method for extracting age-related stereotypes from Twitter data . they generate a corpus of 300,000 over-generalizations about four contemporary generations . |
| Outcome: | The method uncovers common stereotypes as reported in media and psychological literature . it also finds that stereotypes for different generations vary across topics . |
WikiArt Emotions: An Annotated Dataset of Emotions Evoked by Art (L18-1)
Copied to clipboard
| Challenge: | a dataset of 4,000 pieces of art has annotations for emotions evoked in the observer . the dataset can help answer questions about what makes art evocative, how does art convey different emotions, what attributes of a painting make it well liked, and how much does the title impact the affectual response to art. |
| Approach: | They create a dataset of 4,000 western art pieces that has annotations for emotions . they use crowdsourcing to annotate the art for one or more of twenty emotion categories . fear, happiness, love, sadness were the dominant emotions that obtained consistent annotations . |
| Outcome: | The dataset shows that the most popular emotions are fear, happiness, love and sadness . the dataset can be used to develop systems that detect emotions evoked by art . |
Understanding Emotions: A Dataset of Tweets to Study Interactions between Affect Categories (L18-1)
Copied to clipboard
| Challenge: | a new dataset is used to classify text into positive, negative, and neutral classes . a large amount of work on automatic detecting emotions from text has focused on classifying text into basic emotion categories . |
| Approach: | They use Twitter as the source of the textual data they annotate to find out which emotions often present together in tweets . |
| Outcome: | The proposed dataset is useful for training and testing supervised machine learning algorithms . it is based on the results of the SemEval-2018 task 1: Affect in Tweets . |
Quantifying Qualitative Data for Understanding Controversial Issues (L18-1)
Copied to clipboard
| Challenge: | 'Controversy' is a state of sustained public debate on a topic or issue that evokes conflicting opinions, beliefs, claims, arguments, and points of view. |
| Approach: | They propose a crowdsourced approach to quantifying qualitative information on controversial issues by analyzing crowdsourced assertions in social media. |
| Outcome: | The proposed dataset consists of over 2,000 assertions on 16 controversial issues. |
Tackling Social Bias against the Poor: a Dataset and a Taxonomy on Aporophobia (2025.findings-naacl)
Copied to clipboard
Georgina Curto, Svetlana Kiritchenko, Muhammad Hammad Fahim Siddiqui, Isar Nejadgholi, Kathleen C. Fraser
| Challenge: | Poverty is a multidimensional phenomenon that affects 712 million people worldwide . |
| Approach: | They propose to annotate a corpus of English tweets from five world regions for the presence of harmful beliefs and discriminative actions against poor people on social media. |
| Outcome: | The proposed model can be used to identify, track and mitigat aporophobia on social media at scale. |
Big BiRD: A Large, Fine-Grained, Bigram Relatedness Dataset for Examining Semantic Composition (N19-1)
Copied to clipboard
| Challenge: | Existing datasets of semantic relatedness only include pairs of unigrams (single words) Existing data suffer from inconsistent annotations and scale region bias due to rating scales. |
| Approach: | They propose to use a large, fine-grained, bigram relatedness dataset to compare the relatedness of 3,345 English term pairs using a comparative annotation technique called Best–Worst Scaling. |
| Outcome: | The proposed datasets are highly reliable and have a split-half reliability of 0.937. |
Necessity and Sufficiency for Explaining Text Classifiers: A Case Study in Hate Speech Detection (2022.naacl-main)
Copied to clipboard
| Challenge: | XAI features usually provide a single importance score for each token, but feature attribution methods provide two complementary and theoretically-grounded scores for each utterance. |
| Approach: | They propose a feature attribution method that generates explicit perturbations of the input text, allowing the importance scores themselves to be explainable. |
| Outcome: | The proposed method explain the predictions of hate speech detection models on a set of curated examples from a test suite. |
Improving Generalizability in Implicitly Abusive Language Detection with Concept Activation Vectors (2022.acl-long)
Copied to clipboard
| Challenge: | a new study shows that general abusive language classifiers are reliable in detecting explicit abuse but fail to detect more subtle abuses. |
| Approach: | They propose an interpretability technique to quantify the sensitivity of a trained model to new data . they propose a degree of explicitness metric to suggest out-of-domain unlabeled examples . |
| Outcome: | The proposed interpretability technique is useful for predicting the generalizability of the model on new data. |
Challenging Negative Gender Stereotypes: A Study on the Effectiveness of Automated Counter-Stereotypes (2024.lrec-main)
Copied to clipboard
| Challenge: | Gender stereotypes are pervasive beliefs about individuals based on their gender that shape societal attitudes, behaviours, and even opportunities. |
| Approach: | They propose eleven strategies to automatically counteract gender stereotypes by generating gender-based counter-stereotypes from a questionnaire to male and female participants. |
| Outcome: | The proposed strategies were perceived as offensive and/or implausible by the raters . humour, perspective-taking, counter-examples, and empathy for the speaker were perceived to be less effective. |
Uncovering Bias in Large Vision-Language Models at Scale with Counterfactuals (2025.naacl-long)
Copied to clipboard
| Challenge: | Large Vision-Language Models (LVLMs) have been proposed to augment LLMs with visual inputs. |
| Approach: | They propose large vision-Language Models to augment LLMs with visual inputs. |
| Outcome: | The proposed models condition generated text on both an input image and a visual prompt, enabling a variety of use cases such as visual question answering and multimodal chat. |
When Detection Fails: The Power of Fine-Tuned Models to Generate Human-Like Social Media Text (2025.findings-acl)
Copied to clipboard
| Challenge: | detecting AI-generated text on social media is difficult due to short text length and informal language of the internet . a recent study shows that detection of AI-generated posts is difficult under assumptions that an attacker has no knowledge of the generating model . |
| Approach: | They use open-source, closed-source and fine-tuned social media to detect AI-generated text . they use assumptions about knowledge of and access to the generating models to test detection . |
| Outcome: | a human study shows that detection of AI-generated social media posts is difficult . the study compared 505,159 posts from open-source, closed-source and fine-tuned models . |
SOLO: A Corpus of Tweets for Examining the State of Being Alone (2020.lrec-1)
Copied to clipboard
| Challenge: | Psychologists distinguish between the concept of solitude, a positive state of voluntary aloneness, and the concept 'loneliness', characterized as dissatisfaction with the quality of one’s social interactions. |
| Approach: | They present a corpus of over 4 million tweets with query terms solitude, lonely, and loneliness. |
| Outcome: | The proposed analysis analyzes over 4 million tweets with the terms solitude, lonely, and loneliness. |
Examining Gender and Racial Bias in Large Vision–Language Models Using a Novel Dataset of Parallel Images (2024.eacl-long)
Copied to clipboard
| Challenge: | a new wave of large vision–language models (LVLMs) incorporate images as input in addition to text . a recent study examined potential gender and racial biases in such systems based on the perceived characteristics of the people in the input images. |
| Approach: | They examine potential gender and racial biases in large vision–language models . they query a dataset of AI-generated images of people to see whether they differ . |
| Outcome: | The proposed dataset shows that the images differ in gender and race according to the perceived characteristics of the person depicted. |
Understanding and Countering Stereotypes: A Computational Approach to the Stereotype Content Model (2021.acl-long)
Copied to clipboard
| Challenge: | Stereotypical language expresses widely-held beliefs about different social categories. |
| Approach: | They propose a computational approach to interpreting stereotypes in text through the Stereotype Content Model (SCM), a comprehensive causal theory from social psychology. |
| Outcome: | The proposed model compares favourably with survey-based studies in the psychological literature on stereotypes and shows that it is realistic and effective. |
Adaptable Moral Stances of Large Language Models on Sexist Content: Implications for Society and Gender Discourse (2024.emnlp-main)
Copied to clipboard
| Challenge: | Using large language models, large language model learning has become more integrated into our daily lives, making it increasingly important to ensure they reflect ethical and equitable values. |
| Approach: | They assess how LLMs can apply moral reasoning to both criticize and defend sexist language by evaluating their models and evaluating the moral foundations cited by them. |
| Outcome: | The models show they can provide comprehensible and contextually relevant text for understanding diverse views on how sexism is perceived. |