Dependency parsing is the task of inferring natural language structure, often approached by modeling word interactions via attention through biaffine scoring. This mechanism works like self-attention in Transformers, where scores are calculated for every pair of words in a sentence. However, unlike Transformer attention, biaffine scoring does not use normalization prior to taking the softmax of the scores. In this paper, we provide theoretical evidence and empirical results revealing that a lack of normalization necessarily results in overparameterized parser models, where the extra parameters compensate for the sharp softmax outputs produced by high variance inputs to the biaffine scoring function. We argue that biaffine scoring can be made substantially more efficient by performing score normalization. We conduct experiments on semantic and syntactic dependency parsing in multiple languages, along with latent graph inference on non-linguistic data, using various settings of a $k$-hop parser. We train $N$-layer stacked BiLSTMs and evaluate the parser's performance with and without normalizing biaffine scores. Normalizing allows us to achieve state-of-the-art performance with fewer samples and trainable parameters. Code: https://github.com/paolo-gajo/EfficientSDP

Data will be made available on request

Research on knowledge graph construction (KGC) has recently shown great promise also thanks to the adoption of large language models (LLM) for the automatic extraction of structured information from raw text. However, most works rely on commercial, closed-source LLMs, hindering reproducibility and accessibility. We explore KGC with smaller, open-weight LLMs and investigate whether they can be used to improve upon the results obtained by systems leveraging bigger, closed-source models. Specifically, we focus on CodeKGC, a prompting framework based on GPT-3.5. We choose a variety of models either pre-trained primarily on natural language or on code and fine-tune them on three datasets used for information extraction. We fine-tune with prompts formatted either in natural language or as Python-like scripts. In addition, we optionally train the models with prompts including chain-of-thought sections. After fine-tuning, the choice of coding vs natural language prompts has a limited impact on performance, while chain-of-thought training mostly leads to a performance decrease. Moreover, we show that a LLM can be outperformed by much smaller versions on this task, after undergoing the same amount of training. We find that in general the selected lightweight LLMs outperform the much larger CodeKGC by as much as 15–20 absolute F 1 points after fine-tuning. The results show that state-of-the-art KGC systems can be developed using smaller and open-weight models, enhancing research transparency, lowering compute requirements, and decreasing third-party API reliance. Code: https://github.com/TinfFoil/natcode-llm-kgc

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.

Hate speech is infamously one of the most demanding topics in Natural Language Processing, as its multifacetedness is accompanied by a handful of challenges, such as multilinguality and cross-linguality. Hate speech has a subjective aspect that intensifies when referring to different cultures and different languages. In this respect, we design a pipeline that will help us explore the possibility of the creation of a parallel multilingual hate speech dataset, using machine translation. In this paper, we evaluate how/whether this is feasible by assessing the quality of the translations, calculating the toxicity levels of original and target texts, and calculating correlations between the newly obtained scores. Finally, we perform a qualitative analysis to gain further semantic and grammatical insights. With this pipeline we aim at exploring ways of filtering hate speech texts in order to parallelize sentences in multiple languages, examining the challenges of 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.

Cristina España-Bonet, Alberto Barrón-Cedeño. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2024.

We approach the task of assessing the suitability of a source text for translation by transferring the knowledge from established MT evaluation metrics to a model able to predict MT quality a priori from the source text alone. To open the door to experiments in this regard, we depart from reference English-German parallel corpora to build a corpus of 14,253 source text-quality score tuples. The tuples include four state-of-the-art metrics: cushLEPOR, BERTScore, COMET, and TransQuest. With this new resource at hand, we fine-tune XLM-RoBERTa, both in a single-task and a multi-task setting, to predict these evaluation scores from the source text alone. Results for this methodology are promising, with the single-task model able to approximate well-established MT evaluation and quality estimation metrics - without looking at the actual machine translations - achieving low RMSE values in the [0.1-0.2] range and Pearson correlation scores up to 0.688.

Abstract We propose a language-independent graph-based method to build à-la-carte article collections on user-defined domains from the Wikipedia. The core model is based on the exploration of the encyclopedia’s category graph and can produce both mono- and multilingual comparable collections. We run thorough experiments to assess the quality of the obtained corpora in 10 languages and 743 domains. According to an extensive manual evaluation, our graph model reaches an average precision of $$84\%$$ 84 % on in-domain articles, outperforming an alternative model based on information retrieval techniques. As manual evaluations are costly, we introduce the concept of domainness and design several automatic metrics to account for the quality of the collections. Our best metric for domainness shows a strong correlation with human judgments, representing a reasonable automatic alternative to assess the quality of domain-specific corpora. We release the toolkit with the implementation of the extraction methods, the evaluation measures and several utilities.

Some human preferences are universal.The odor of vanilla is perceived as pleasant all around the world.We expect neural models trained on human texts to exhibit these kind of preferences, i.e. biases, but we show that this is not always the case.We explore 16 static and contextual embedding models in 9 languages and, when possible, compare them under similar training conditions.We introduce and release CA-WEAT, multilingual cultural aware tests to quantify biases, and compare them to previous English-centric tests.Our experiments confirm that monolingual static embeddings do exhibit human biases, but values differ across languages, being far from universal.Biases are less evident in contextual models, to the point that the original human association might be reversed.Multilinguality proves to be another variable that attenuates and even reverses the effect of the bias, specially in contextual multilingual models.In order to explain this variance among models and languages, we examine the effect of asymmetries in the training corpus, departures from isomorphism in multilingual embedding spaces and discrepancies in the testing measures between languages.

We target the complementary binary tasks of identifying whether a tweet is misogynous and, if that is the case, whether it is also aggressive. We compare two ways to address these problems: one multi-class model that discriminates between all the classes at once: not misogynous, non aggressive-misogynous and aggressive-misogynous; as well as a cascaded approach where the binary classification is carried out separately (misogynous vs non-misogynous and aggressive vs non-aggressive) and then joined together. For the latter, two training and three testing scenarios are considered. Our models are built on top of AlBERTo and are evaluated on the framework of Evalita’s 2020 shared task on automatic misogyny and aggressiveness identification in Italian tweets. Our cascaded models —including the strong naïve baseline— outperform significantly the top submissions to Evalita, reaching state-of-the-art performance without relying on any external information.

The reporting and the analysis of current events around the globe has expanded from professional, editor-lead journalism all the way to citizen journalism. Nowadays, politicians and other key players enjoy direct access to their audiences through social media, bypassing the filters of official cables or traditional media. However, the multiple advantages of free speech and direct communication are dimmed by the misuse of media to spread inaccurate or misleading claims. These phenomena have led to the modern incarnation of the fact-checker --- a professional whose main aim is to examine claims using available evidence and to assess their veracity. Here, we survey the available intelligent technologies that can support the human expert in the different steps of her fact-checking endeavor. These include identifying claims worth fact-checking, detecting relevant previously fact-checked claims, retrieving relevant evidence to fact-check a claim, and actually verifying a claim. In each case, we pay attention to the challenges and the potential impact on real-world fact-checking.

Propaganda campaigns aim at influencing people's mindset with the purpose of advancing a specific agenda. They exploit the anonymity of the Internet, the micro-profiling ability of social networks, and the ease of automatically creating and managing coordinated networks of accounts, to reach millions of social network users with persuasive messages, specifically targeted to topics each individual user is sensitive to, and ultimately influencing the outcome on a targeted issue. In this survey, we review the state of the art on computational propaganda detection from the perspective of Natural Language Processing and Network Analysis, arguing about the need for combined efforts between these communities. We further discuss current challenges and future research directions.

We present proppy, the first publicly available real-world, real-time propaganda detection system for online news, which aims at raising awareness, thus potentially limiting the impact of propaganda and helping fight disinformation. The system constantly monitors a number of news sources, deduplicates and clusters the news into events, and organizes the articles about an event on the basis of the likelihood that they contain propagandistic content. The system is trained on known propaganda sources using a variety of stylistic features. The evaluation results on a standard dataset show stateof-the-art results for propaganda detection.

Giovanni Da San Martino, Seunghak Yu, Alberto Barrón-Cedeño, Rostislav Petrov, Preslav Nakov. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019.

The journalism industry has undergone a revolution in the past decade, leading to new opportunities as well as challenges. News consumption, production and delivery have all been affected and transformed by technology Readers require new mechanisms to cope with the vast volume of information in order to be informed about news events. Reporters have begun to use natural language processing (NLP) and (IR) techniques for investigative work. Publishers and aggregators are seeking new business models, and new ways to reach and retain their audience. A shift in business models has led to a gradual shift in styles of journalism in attempts to increase page views; and, far more concerning, to real mis- and dis-information, alongside allegations of "fake news" threatening the journalistic freedom and integrity of legitimate news outlets. Social media platforms drive viewership, creating filter bubbles and an increasingly polarized readership. News documents have always been a part of research on information access and retrieval methods. Over the last few years, the IR community has increasingly recognized these challenges in journalism and opened a conversation about how we might begin to address them. Evidence of this recognition is the participation in the two previous editions of our NewsIR workshop, held in ECIR 2016 and 2018. One of the most important outcomes of those workshops is an increasing awareness in the community about the changing nature of journalism and the IR challenges it entails. To move yet another step forward, the goal of the third edition of our workshop will be to create a multidisciplinary venue that brings together news experts from both technology and journalism. This would take NewsIR from a European forum targeting mainly IR researchers, into a more inclusive and influential international forum. We hope that this new format will foster further understanding for both news professionals and IR researchers, as well as producing better outcomes for news consumers. We will address the possibilities and challenges that technology offers to the journalists, the challenges that new developments in journalism create for IR researchers, and the complexity of information access tasks for news readers.

Community Question Answering (cQA) forums are very popular nowadays, as they represent effective means for communities around particular topics to share information. Unfortunately, this information is not always factual. Thus, here we explore a new dimension in the context of cQA, which has been ignored so far: checking the veracity of answers to particular questions in cQA forums. As this is a new problem, we create a specialized dataset for it. We further propose a novel multi-faceted model, which captures information from the answer content (what is said and how), from the author profile (who says it), from the rest of the community forum (where it is said), and from external authoritative sources of information (external support). Evaluation results show a MAP value of 86.54, which is 21 points absolute above the baseline.

In many Information Retrieval tasks, the boundary between classes is not well defined, and assigning a document to a specific class may be complicated, even for humans. For instance, a document which is not directly related to the user's query may still contain relevant information. In this scenario, an option is to define an intermediate class collecting ambiguous instances. Yet some natural questions arise. Is this annotation strategy convenient? how should the intermediate class be treated? To answer these questions, we explored two community question answering datasets whose comments were originally annotated with three classes. We re-annotated a subset of instances considering a binary good vs bad setting. Our main contribution is to show empirically that the inclusion of an intermediate class to assess Boolean relevance is not useful. Moreover, in case the data is already annotated with a 3-class strategy, the instances from the intermediate class can be safely removed at training time.

We study how to find relevant questions in community forums when the language of the new questions is different from that of the existing questions in the forum. In particular, we explore the Arabic-English language pair. We compare a kernel-based system with a feed-forward neural network in a scenario where a large parallel corpus is available for training a machine translation system, bilingual dictionaries, and cross-language word embeddings. We observe that both approaches degrade the performance of the system when working on the translated text, especially the kernel-based system, which depends heavily on a syntactic kernel. We address this issue using a cross-language tree kernel, which compares the original Arabic tree to the English trees of the related questions. We show that this kernel almost closes the performance gap with respect to the monolingual system. On the neural network side, we use the parallel corpus to train cross-language embeddings, which we then use to represent the Arabic input and the English related questions in the same space. The results also improve to close to those of the monolingual neural network. Overall, the kernel system shows a better performance compared to the neural network in all cases.

We study the impact of different types of features for question ranking in community Question Answering: bag-of-words models (BoW), syntactic tree kernels (TKs) and rank features. It should be noted that structural kernels have never been applied to the question reranking task, i.e., question to question similarity, where they have to model paraphrase relations. Additionally, the informal text, typically present in forums, poses new challenges to the use of TKs. We compare our learning to rank (L2R) algorithms against a strong baseline given by the Google rank (GR). The results show that (i) our shallow structures used in TKs are robust enough to noisy data and (ii) improving GR requires effective BoW features and TKs along with an accurate model of GR features in the used L2R algorithm.

Community question answering (cQA) websites are focused on users who query questions onto an online forum, expecting for other users to provide them answers or suggestions. Unlike other social media, the length of the posted queries has no limits and queries tend to be multi-sentence elaborations combining context, actual questions, and irrelevant information. We approach the problem of question ranking: given a user’s new question, to retrieve those previously-posted questions which could be equivalent, or highly relevant. This could prevent the posting of nearly-duplicate questions and provide the user with instantaneous answers. For the first time in cQA, we address the selection of relevant text —both at sentence- and at constituent-level— for parse tree-based representations. Our supervised models for text selection boost the performance of a tree kernel-based machine learning model, allowing it to overtake the current state of the art on a recently released cQA evaluation framework.

In real-world data, e.g., from Web forums, text is often contaminated with redundant or irrelevant content, which leads to introducing noise in machine learning algorithms. In this paper, we apply Long Short-Term Memory networks with an attention mechanism, which can select important parts of text for the task of similar question retrieval from community Question Answering (cQA) forums. In particular, we use the attention weights for both selecting entire sentences and their subparts, i.e., word/chunk, from shallow syntactic trees. More interestingly, we apply tree kernels to the filtered text representations, thus exploiting the implicit features of the subtree space for learning question reranking. Our results show that the attention-based pruning allows for achieving the top position in the cQA challenge of SemEval 2016, with a relatively large gap from the other participants while greatly decreasing running time.

Shafiq Joty, Alberto Barrón-Cedeño, Giovanni Da San Martino, Simone Filice, Lluís Màrquez, Alessandro Moschitti, Preslav Nakov. Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing. 2015.

In this article we present a three-step methodology for dynamically improving a statistical machine translation (SMT) system by incorporating human feedback in the form of free edits on the system translations. We target at feedback provided by casual users, which is typically error-prone. Thus, we first propose a filtering step to automatically identify the better user-edited translations and discard the useless ones. A second step produces a pivot-based alignment between source and user-edited sentences, focusing on the errors made by the system. Finally, a third step produces a new translation model and combines it linearly with the one from the original system. We perform a thorough evaluation on a real-world dataset collected from the Reverso.net translation service and show that every step in our methodology contributes significantly to improve a general purpose SMT system. Interestingly, the quality improvement is not only due to the increase of lexical coverage, but to a better lexical selection, reordering, and morphology. Finally, we show the robustness of the methodology by applying it to a different scenario, in which the new examples come from an automatically Web-crawled parallel corpus. Using exactly the same architecture and models provides again a significant improvement of the translation quality of a general purpose baseline SMT system.

Although paraphrasing is the linguistic mechanism underlying many plagiarism cases, little attention has been paid to its analysis in the framework of automatic plagiarism detection. Therefore, state-of-the-art plagiarism detectors find it difficult to detect cases of paraphrase plagiarism. In this article, we analyze the relationship between paraphrasing and plagiarism, paying special attention to which paraphrase phenomena underlie acts of plagiarism and which of them are detected by plagiarism detection systems. With this aim in mind, we created the P4P corpus, a new resource that uses a paraphrase typology to annotate a subset of the PAN-PC-10 corpus for automatic plagiarism detection. The results of the Second International Competition on Plagiarism Detection were analyzed in the light of this annotation. The presented experiments show that (i) more complex paraphrase phenomena and a high density of paraphrase mechanisms make plagiarism detection more difficult, (ii) lexical substitutions are the paraphrase mechanisms used the most when plagiarizing, and (iii) paraphrase mechanisms tend to shorten the plagiarized text. For the first time, the paraphrase mechanisms behind plagiarism have been analyzed, providing critical insights for the improvement of automatic plagiarism detection systems.

Plagiarism, the unacknowledged reuse of text, has increased in recent years due to the large amount of texts readily available. For instance, recent studies claim that nowadays a high rate of student reports include plagiarism, making manual plagiarism detection practically infeasible.