Warning: This paper contains explicit statements of offensive stereotypes which may be upsetting The study of bias, fairness and social impact in Natural Language Processing (NLP) lacks resources in languages other than English. Our objective is to support the evaluation of bias in language models in a multilingual setting. We use stereotypes across nine types of biases to build a corpus containing contrasting sentence pairs, one sentence that presents a stereotype concerning an underadvantaged group and another minimally changed sentence, concerning a matching advantaged group. We build on the French CrowS-Pairs corpus and guidelines to provide translations of the existing material into seven additional languages. In total, we produce 11,139 new sentence pairs that cover stereotypes dealing with nine types of biases in seven cultural contexts. We use the final resource for the evaluation of relevant monolingual and multilingual masked language models. We find that language models in all languages favor sentences that express stereotypes in most bias categories. The process of creating a resource that covers a wide range of language types and cultural settings highlights the difficulty of bias evaluation, in particular comparability across languages and contexts.

Visual Question Answering~(VQA) is an important task in multimodal AI, which requires models to understand and reason on knowledge present in visual and textual data. However, most of the current VQA datasets and models are primarily focused on English and a few major world languages, with images that are Western-centric. While recent efforts have tried to increase the number of languages covered on VQA datasets, they still lack diversity in low-resource languages. More importantly, some datasets extend the text to other languages, either via translation or some other approaches, but usually keep the same images, resulting in narrow cultural representation. To address these limitations, we create CVQA, a new Culturally-diverse Multilingual Visual Question Answering benchmark dataset, designed to cover a rich set of languages and regions, where we engage native speakers and cultural experts in the data collection process. CVQA includes culturally-driven images and questions from across 28 countries in four continents, covering 26 languages with 11 scripts, providing a total of 9k questions. We benchmark several Multimodal Large Language Models (MLLMs) on CVQA, and we show that the dataset is challenging for the current state-of-the-art models. This benchmark will serve as a probing evaluation suite for assessing the cultural bias of multimodal models and hopefully encourage more research efforts towards increasing cultural awareness and linguistic diversity in this field.

Since BERT appeared, Transformer language models and transfer learning have become state-of-the-art for natural language processing tasks. Recently, some works geared towards pre-training specially-crafted models for particular domains, such as scientific papers, medical documents, user-generated texts, among others. These domain-specific models have been shown to improve performance significantly in most tasks; however, for languages other than English, such models are not widely available. In this work, we present RoBERTuito, a pre-trained language model for user-generated text in Spanish, trained on over 500 million tweets. Experiments on a benchmark of tasks involving user-generated text showed that RoBERTuito outperformed other pre-trained language models in Spanish. In addition to this, our model has some cross-lingual abilities, achieving top results for English-Spanish tasks of the Linguistic Code-Switching Evaluation benchmark (LinCE) and also competitive performance against monolingual models in English Twitter tasks. To facilitate further research, we make RoBERTuito publicly available at the HuggingFace model hub together with the dataset used to pre-train it.

In this paper we present a bootstrapping approach for attribute value extraction that minimizes the need for human intervention. Our approach automatically extracts attribute names and values from semi-structured text, generates a small labelled dataset, and bootstraps it by extracting new values from unstructured text. It is domain/language-independent, relying only on existing semi-structured text to create the initial labeled dataset. We assess the impact of different machine learning approaches to increase precision of the core approach without compromising coverage. We perform an extensive evaluation using e-commerce product data across different categories in two languages and hundreds of thousands of product pages. We show that our approach provides high precision and good coverage. In addition, we study the impact of different methods that address specific sources of error. With error analysis we highlight how these methods complement each other, obtaining insights about the individual methods and the ensamble as a whole.

We present a neural architecture to model student behavior in virtual educational environments using purely unsupervised information. A crucial part of this architecture is the optimization of a joint embedding function to represent both students and course elements into a single shared space. This joint representation is more adequate than disjoint representations because it elicits insights on the relations between students and contents. Moreover, the model is trained only with interactions of the student with online learning platforms, without requiring any additional manual labeling by experts. We obtain state-of-the-art results using this approach in two types of task: first, dropout prediction in online courses (MOOCs), and second Knowledge Tracing in Intelligent Tutoring Systems (ITS). We explore how the deep architecture is flexible enough to capture variables related to different phenomena, such as engagement or skill mastery.

In this paper we try to improve Information Extraction in legal texts by creating a legal Named Entity Recognizer, Classifier and Linker. With this tool, we can identify relevant parts of texts and connect them to a structured knowledge representation, the LKIF ontology.

Managing licensing information and data rights is becoming a crucial issue in the Linked (Open) Data scenario. An open problem in this scenario is how to associate machine-readable licenses specifications to the data, so that automated approaches to treat such information can be fruitfully exploited to avoid data misuse. This means that we need a way to automatically extract from a natural language document specifying a certain license a machine-readable description of the terms of use and reuse identified in such license.