In this paper, we apply a method to quantify biases associated with named entities from various countries. We create counterfactual examples with small perturbations on target-domain data instead of relying on templates or specific datasets for bias detection. On widely used classifiers for subjectivity analysis, including sentiment, emotion, hate speech, and offensive text using Twitter data, our results demonstrate positive biases related to the language spoken in a country across all classifiers studied. Notably, the presence of certain country names in a sentence can strongly influence predictions, up to a 23% change in hate speech detection and up to a 60% change in the prediction of negative emotions such as anger. We hypothesize that these biases stem from the training data of pre-trained language models (PLMs) and find correlations between affect predictions and PLMs likelihood in English and unknown languages like Basque and Maori, revealing distinct patterns with exacerbate correlations. Further, we followed these correlations in-between counterfactual examples from a same sentence to remove the syntactical component, uncovering interesting results suggesting the impact of the pre-training data was more important for English-speaking-country names.

This paper presents the results of the WASSA 2024 shared task on predicting empathy, emotion, and personality in conversations and reactions to news articles. Participating teams were given access to a new, unpublished extension of the WASSA 2023 shared task dataset. This task is both multi-level and multi-modal: data is available at the person, essay, dialog, and dialog-turn levels and includes formal (news articles) and informal text (essays and dialogs), self-report data (personality and distress), and third-party annotations (empathy and emotion). The shared task included a new focus on conversations between humans and LLM-based virtual agents which occur immediately after reading and reacting to the news articles. Participants were encouraged to explore the multi-level and multi-modal nature of this data. Participation was encouraged in four tracks: (i) predicting the perceived empathy at the dialog level, (ii) predicting turn-level empathy, emotion polarity, and emotion intensity in conversations, (iii) predicting state empathy and distress scores, and (iv) predicting personality. In total, 14 teams participated in the shared task. We summarize the methods and resources used by the participating teams.

Classical bias detection methods used in Machine Learning are themselves biased because of the different confounding variables implied in the assessment of the initial biases. First they are using templates that are syntactically simple and distant from the target data on which the model will deployed. Second, current methods are assessing biases in pre-trained language models or in dataset, but not directly on the fine-tuned classifier that can actually produce harms. We propose a simple method to detect the biases of a specific fine-tuned classifier on any type of unlabeled data. The idea is to study the classifier behavior by creating counterfactual examples directly on the target data distribution and quantify the amount of changes. In this work, we focus on named entity perturbations by applying a Named Entity Recognition on target-domain data and modifying them accordingly to most common names or location of a target group (gender and country), and this for several morphosynctactically different languages spoken in relation with the countries of the target groups. We used our method on two models available open-source that are likely to be deployed by industry, and on two tasks and domains. We first assess the bias of a multilingual sentiment analysis model trained over multiple-languages tweets and available open-source, and then a multilingual stance recognition model trained over several languages and assessed over English language. Finally we propose to link the perplexity of each example with the bias of the model, by looking at the change in label distribution with respect to the language of the target group. Our work offers a fine-grained analysis of the interactions between names and languages, revealing significant biases in multilingual models.

While human values play a crucial role in making arguments persuasive, we currently lack the necessary extensive datasets to develop methods for analyzing the values underlying these arguments on a large scale. To address this gap, we present the Touché23-ValueEval dataset, an expansion of the Webis-ArgValues-22 dataset. We collected and annotated an additional 4780 new arguments, doubling the dataset’s size to 9324 arguments. These arguments were sourced from six diverse sources, covering religious texts, community discussions, free-text arguments, newspaper editorials, and political debates. Each argument is annotated by three crowdworkers for 54 human values, following the methodology established in the original dataset. The Touché23-ValueEval dataset was utilized in the SemEval 2023 Task 4. ValueEval: Identification of Human Values behind Arguments, where an ensemble of transformer models demonstrated state-of-the-art performance. Furthermore, our experiments show that a fine-tuned large language model, Llama-2-7B, achieves comparable results.

We propose a general method to break down a main complex task into a set of intermediary easier sub-tasks, which are formulated in natural language as binary questions related to the final target task. Our method allows for representing each example by a vector consisting of the answers to these questions. We call this representation Natural Language Learned Features (NLLF). NLLF is generated by a small transformer language model (e.g., BERT) that has been trained in a Natural Language Inference (NLI) fashion, using weak labels automatically obtained from a Large Language Model (LLM). We show that the LLM normally struggles for the main task using in-context learning, but can handle these easiest subtasks and produce useful weak labels to train a BERT. The NLI-like training of the BERT allows for tackling zero-shot inference with any binary question, and not necessarily the ones seen during the training. We show that this NLLF vector not only helps to reach better performances by enhancing any classifier, but that it can be used as input of an easy-to-interpret machine learning model like a decision tree. This decision tree is interpretable but also reaches high performances, surpassing those of a pre-trained transformer in some cases. We have successfully applied this method to two completely different tasks: detecting incoherence in students’ answers to open-ended mathematics exam questions, and screening abstracts for a systematic literature review of scientific papers on climate change and agroecology.

This paper presents the results of the WASSA 2023 shared task on predicting empathy, emotion, and personality in conversations and reactions to news articles. Participating teams were given access to a new dataset from Omitaomu et al. (2022) comprising empathic and emotional reactions to news articles. The dataset included formal and informal text, self-report data, and third-party annotations. Specifically, the dataset contained news articles (where harm is done to a person, group, or other) and crowd-sourced essays written in reaction to the article. After reacting via essays, crowd workers engaged in conversations about the news articles. Finally, the crowd workers self-reported their empathic concern and distress, personality (using the Big Five), and multi-dimensional empathy (via the Interpersonal Reactivity Index). A third-party annotated both the conversational turns (for empathy, emotion polarity, and emotion intensity) and essays (for multi-label emotions). Thus, the dataset contained outcomes (self-reported or third-party annotated) at the turn level (within conversations) and the essay level. Participation was encouraged in five tracks: (i) predicting turn-level empathy, emotion polarity, and emotion intensity in conversations, (ii) predicting state empathy and distress scores, (iii) predicting emotion categories, (iv) predicting personality, and (v) predicting multi-dimensional trait empathy. In total, 21 teams participated in the shared task. We summarize the methods and resources used by the participating teams.

Artificial neural networks typically struggle in generalizing to out-of-context examples. One reason for this limitation is caused by having datasets that incorporate only partial information regarding the potential correlational structure of the world. In this work, we propose TIDA (Targeted Image-editing Data Augmentation), a targeted data augmentation method focused on improving models’ human-like abilities (e.g., gender recognition) by filling the correlational structure gap using a text-to-image generative model. More specifically, TIDA identifies specific skills in captions describing images (e.g., the presence of a specific gender in the image), changes the caption (e.g., “woman” to “man”), and then uses a text-to-image model to edit the image in order to match the novel caption (e.g., uniquely changing a woman to a man while maintaining the context identical). Based on the Flickr30K benchmark, we show that, compared with the original data set, a TIDA-enhanced dataset related to gender, color, and counting abilities induces better performance in several image captioning metrics. Furthermore, on top of relying on the classical BLEU metric, we conduct a fine-grained analysis of the improvements of our models against the baseline in different ways. We compared text-to-image generative models and found different behaviors of the image captioning models in terms of encoding visual encoding and textual decoding.

In this paper, we present the process we used in order to collect new annotations of opinions over the multimodal corpus SEMAINE composed of dyadic interactions. The dataset had already been annotated continuously in two affective dimensions related to the emotions: Valence and Arousal. We annotated the part of SEMAINE called Solid SAL composed of 79 interactions between a user and an operator playing the role of a virtual agent designed to engage a person in a sustained, emotionally colored conversation. We aligned the audio at the word level using the available high-quality manual transcriptions. The annotated dataset contains 5627 speech turns for a total of 73,944 words, corresponding to 6 hours 20 minutes of dyadic interactions. Each interaction has been labeled by three annotators at the speech turn level following a three-step process. This method allows us to obtain a precise annotation regarding the opinion of a speaker. We obtain thus a dataset dense in opinions, with more than 48% of the annotated speech turns containing at least one opinion. We then propose a new baseline for the detection of opinions in interactions improving slightly a state of the art model with RoBERTa embeddings. The obtained results on the database are promising with a F1-score at 0.72.

Stance Recognition over proposals is the task of automatically detecting whether a comment on a specific proposal is in favor of this proposal, against this proposal or that neither inference is likely. The dataset that we propose to use is an online debating platform inaugurated in 2021, where users can submit proposals and comment over proposals or over other comments. It contains 4.2k proposals and 20k comments focused on various topics. Every comment and proposal can come written in another language, with more than 40% of the proposal/comment pairs containing at least two languages, creating a unique intra-multilingual setting. A portion of the data (more than 7k comment/proposal pairs, in 26 languages) was annotated by the writers with a self-tag assessing whether they are in favor or against the proposal. Another part of the data (without self-tag) has been manually annotated: 1206 comments in 6 morphologically different languages (fr, de, en, el, it, hu) were tagged, leading to a Krippendorff’s α of 0.69. This setting allows defining an intra-multilingual and multi-target stance classification task over online debates.

This paper presents the results that were obtained from WASSA 2022 shared task on predicting empathy, emotion, and personality in reaction to news stories. Participants were given access to a dataset comprising empathic reactions to news stories where harm is done to a person, group, or other. These reactions consist of essays and Batson’s empathic concern and personal distress scores. The dataset was further extended in WASSA 2021 shared task to include news articles, person-level demographic information (e.g. age, gender), personality information, and Ekman’s six basic emotions at essay level Participation was encouraged in four tracks: predicting empathy and distress scores, predicting emotion categories, predicting personality and predicting interpersonal reactivity. In total, 14 teams participated in the shared task. We summarize the methods and resources used by the participating teams.

We present a new dataset of online debates in English, annotated with stance. The dataset was scraped from the “Debating Europe” platform, where users exchange opinions over different subjects related to the European Union. The dataset is composed of 2600 comments pertaining to 18 debates related to the “European Green Deal”, in a conversational setting. After presenting the dataset and the annotated sub-part, we pre-train a model for a multilingual stance classification over the X-stance dataset before fine-tuning it over our dataset, and vice-versa. The fine-tuned models are shown to improve stance classification performance on each of the datasets, even though they have different languages, topics and targets. Subsequently, we propose to enhance the performances over “Debating Europe” with an interaction-aware model, taking advantage of the online debate structure of the platform. We also propose a semi-supervised self-training method to take advantage of the imbalanced and unlabeled data from the whole website, leading to a final improvement of accuracy by 3.4% over a Vanilla XLM-R model.

This paper presents the results that were obtained from the WASSA 2021 shared task on predicting empathy and emotions. The participants were given access to a dataset comprising empathic reactions to news stories where harm is done to a person, group, or other. These reactions consist of essays, Batson empathic concern, and personal distress scores, and the dataset was further extended with news articles, person-level demographic information (age, gender, ethnicity, income, education level), and personality information. Additionally, emotion labels, namely Ekman’s six basic emotions, were added to the essays at both the document and sentence level. Participation was encouraged in two tracks: predicting empathy and predicting emotion categories. In total five teams participated in the shared task. We summarize the methods and resources used by the participating teams.

Tweets are specific text data when compared to general text. Although sentiment analysis over tweets has become very popular in the last decade for English, it is still difficult to find huge annotated corpora for non-English languages. The recent rise of the transformer models in Natural Language Processing allows to achieve unparalleled performances in many tasks, but these models need a consequent quantity of text to adapt to the tweet domain. We propose the use of a multilingual transformer model, that we pre-train over English tweets on which we apply data-augmentation using automatic translation to adapt the model to non-English languages. Our experiments in French, Spanish, German and Italian suggest that the proposed technique is an efficient way to improve the results of the transformers over small corpora of tweets in a non-English language.

In this paper we present a new method to learn a model robust to typos for a Named Entity Recognition task. Our improvement over existing methods helps the model to take into account the context of the sentence inside a justice decision in order to recognize an entity with a typo. We used state-of-the-art models and enriched the last layer of the neural network with high-level information linked with the potential of the word to be a certain type of entity. More precisely, we utilized the similarities between the word and the potential entity candidates the tagged sentence context. The experiments on a dataset of french justice decisions show a reduction of the relative F1-score error of 32%, upgrading the score obtained with the most competitive fine-tuned state-of-the-art system from 94.85% to 96.52%.