A popular end-to-end architecture for selective rationalization is the select-then-predict pipeline, comprising a generator to extract highlights fed to a predictor. Such a cooperative system suffers from suboptimal equilibrium minima due to the dominance of one of the two modules, a phenomenon known as interlocking. While several contributions aimed at addressing interlocking, they only mitigate its effect, often by introducing feature-based heuristics, sampling, and ad-hoc regularizations. We present GenSPP, the first interlocking-free architecture for selective rationalization that does not require any learning overhead, as the above-mentioned. GenSPP avoids interlocking by performing disjoint training of the generator and predictor via genetic global search. Experiments on a synthetic and a real-world benchmark show that our model outperforms several state-of-the-art competitors.

Current research in machine learning and artificial intelligence is largely centered on modeling and performance evaluation, less so on data collection. However, recent research demonstrated that limitations and biases in data may negatively impact trustworthiness and reliability. These aspects are particularly impactful on sensitive domains such as mental health and neurological disorders, where speech data are used to develop AI applications for patients and healthcare providers. In this paper, we chart the landscape of available speech datasets for this domain, to highlight possible pitfalls and opportunities for improvement and promote fairness and diversity. We present a comprehensive list of desiderata for building speech datasets for mental health and neurological disorders and distill it into an actionable checklist focused on ethical concerns to foster more responsible research.

We develop novel annotation guidelines for sentence-level subjectivity detection, which are not limited to language-specific cues. We use our guidelines to collect NewsSD-ENG, a corpus of 638 objective and 411 subjective sentences extracted from English news articles on controversial topics. Our corpus paves the way for subjectivity detection in English and across other languages without relying on language-specific tools, such as lexicons or machine translation. We evaluate state-of-the-art multilingual transformer-based models on the task in mono-, multi-, and cross-language settings. For this purpose, we re-annotate an existing Italian corpus. We observe that models trained in the multilingual setting achieve the best performance on the task.

Misogyny is often expressed through figurative language. Some neutral words can assume a negative connotation when functioning as pejorative epithets. Disambiguating the meaning of such terms might help the detection of misogyny. In order to address such task, we present PejorativITy, a novel corpus of 1,200 manually annotated Italian tweets for pejorative language at the word level and misogyny at the sentence level. We evaluate the impact of injecting information about disambiguated words into a model targeting misogyny detection. In particular, we explore two different approaches for injection: concatenation of pejorative information and substitution of ambiguous words with univocal terms. Our experimental results, both on our corpus and on two popular benchmarks on Italian tweets, show that both approaches lead to a major classification improvement, indicating that word sense disambiguation is a promising preliminary step for misogyny detection. Furthermore, we investigate LLMs’ understanding of pejorative epithets by means of contextual word embeddings analysis and prompting.

We propose misogyny detection as an Argumentative Reasoning task and we investigate the capacity of large language models (LLMs) to understand the implicit reasoning used to convey misogyny in both Italian and English.The central aim is to generate the missing reasoning link between a message and the implied meanings encoding the misogyny.Our study uses argumentation theory as a foundation to form a collection of prompts in both zero-shot and few-shot settings.These prompts integrate different techniques, including chainof-thought reasoning and augmented knowledge.Our findings show that LLMs fall short on reasoning capabilities about misogynistic comments relying on their implicit knowledge derived from internalized common stereotypes about women to generate implied assumptions, rather than on inductive reasoning.

With recent advances in question-answering models, various datasets have been collected to improve and study the effectiveness of these models on scientific texts.Questions and answers in these datasets explore a scientific paper by seeking factual information from the paper's content.However, these datasets do not tackle the argumentative content of scientific papers, which is of huge importance in persuasiveness of a scientific discussion.We introduce ArgSciChat, a dataset of 41 argumentative dialogues between scientists on 20 NLP papers.The unique property of our dataset is that it includes both exploratory and argumentative questions and answers in a dialogue discourse on a scientific paper.Moreover, the size of ArgSciChat demonstrates the difficulties in collecting dialogues for specialized domains.Thus, our dataset is a challenging resource to evaluate dialogue agents in low-resource domains, in which collecting training data is costly.We annotate all sentences of dialogues in ArgSciChat and analyze them extensively.The results confirm that dialogues in ArgSci-Chat include exploratory and argumentative interactions.Furthermore, we use our dataset to fine-tune and evaluate a pre-trained documentgrounded dialogue agent.The agent achieves a low performance on our dataset, motivating a need for dialogue agents with a capability to reason and argue about their answers.We publicly release ArgSciChat 1 .

Argument structure prediction aims to identify the relations between arguments or between parts of arguments. It is a crucial task in legal argument mining, where it could help identifying motivations behind judgments or even fallacies or inconsistencies. It is also a very challenging task, which is relatively underdeveloped compared to other argument mining tasks, owing to a number of reasons including a low availability of datasets and a high complexity of the reasoning involved. In this work, we address argumentative link prediction in decisions by Court of Justice of the European Union on fiscal state aid. We study how propositions are combined in higher-level structures and how the relations between propositions can be predicted by NLP models. To this end, we present a novel annotation scheme and use it to extend a dataset from literature with an additional annotation layer. We use our new dataset to run an empirical study, where we compare two architectures and explore different combinations of hyperparameters and training regimes. Our results indicate that an ensemble of residual networks yields the best results.

AMICA is an argument mining-based search engine, specifically designed for the analysis of scientific literature related to Covid-19. AMICA retrieves scientific papers based on matching keywords and ranks the results based on the papers' argumentative content. An experimental evaluation conducted on a case study in collaboration with the Italian National Institute of Health shows that the AMICA ranking agrees with expert opinion, as well as, importantly, with the impartial quality criteria indicated by Cochrane Systematic Reviews.

This study aims at predicting the outcomes of legal cases based on the textual content of judicial decisions. We present a new corpus of Italian documents, consisting of 226 annotated decisions on Value Added Tax by Regional Tax law commissions. We address the task of predicting whether a request is upheld or rejected in the final decision. We employ traditional classifiers and NLP methods to assess which parts of the decision are more informative for the task.

Consumer contracts often contain unfair clauses, in apparent violation of the relevant legislation.In this paper we present a new methodology for evaluating such clauses in online Terms of Services.We expand a set of tagged documents (terms of service), with a structured corpus where unfair clauses are liked to a knowledge base of rationales for unfairness, and experiment with machine learning methods on this expanded training set.Our experimental study is based on deep neural networks that aim to combine learning and reasoning tasks, one major example being Memory Networks.Preliminary results show that this approach may not only provide reasons and explanations to the user, but also enhance the automated detection of unfair clauses.