Neural end-to-end surface realizers output more fluent texts than classical architectures. However, they tend to suffer from adequacy problems, in particular hallucinations in numerical referring expression generation. This poses a problem to language generation in sensitive domains, as is the case of robot journalism covering COVID-19 and Amazon deforestation. We propose an approach whereby numerical referring expressions are converted from digits to plain word form descriptions prior to being fed to state-of-the-art Large Language Models. We conduct automatic and human evaluations to report the best strategy to numerical superficial realization. Code and data are publicly available.

Previous studies have highlighted the advantages of pipeline neural architectures over end-to-end models, particularly in reducing text hallucination. In this study, we extend prior research by integrating pretrained language models (PLMs) into a pipeline framework, using both fine-tuning and prompting methods. Our findings show that fine-tuned PLMs consistently generate high quality text, especially within end-to-end architectures and at intermediate stages of the pipeline across various domains. These models also outperform prompt-based ones on automatic evaluation metrics but lag in human evaluations. Compared to the standard five-stage pipeline architecture, a streamlined three-stage pipeline, which only include ordering, structuring, and surface realization, achieves superior performance in fluency and semantic adequacy according to the human evaluation.

In this paper, we present our approach to the GEM Shared Task at the INLG’24 Generation Challenges, which focuses on generating data-to-text in multiple languages, including low-resource languages, from WebNLG triples. We employ a combination of end-to-end and pipeline neural architectures for English text generation. To extend our methodology to Hindi, Korean, Arabic, and Swahili, we leverage a neural machine translation model. Our results demonstrate that our approach achieves competitive performance in the given task.

In the digital age, cyberbullying (CB) poses a significant concern, impacting individuals as early as primary school and leading to severe or lasting consequences, including an increased risk of self-harm. CB incidents, are not limited to bullies and victims, but include bystanders with various roles, and usually have numerous sub-categories and variations of online harms. This position paper emphasises the complexity of CB incidents by drawing on insights from psychology, social sciences, and computational linguistics. While awareness of CB complexities is growing, existing computational techniques tend to oversimplify CB as a binary classification task, often relying on training datasets that capture peripheries of CB behaviours. Inconsistent definitions and categories of CB-related online harms across various platforms further complicates the issue. Ethical concerns arise when CB research involves children to role-play CB incidents to curate datasets. Through multi-disciplinary collaboration, we propose strategies for consideration when developing CB detection systems. We present our position on leveraging large language models (LLMs) such as Claude-2 and Llama2-Chat as an alternative approach to generate CB-related role-playing datasets. Our goal is to assist researchers, policymakers, and online platforms in making informed decisions regarding the automation of CB incident detection and intervention. By addressing these complexities, our research contributes to a more nuanced and effective approach to combating CB especially in young people.

We conduct a comparison of pre-trained encoder-only and decoder-only language models with and without continued pre-training, to detect online sexism. Our fine-tuning-based classifier system achieved the 16th rank in the SemEval 2023 Shared Task 10 Subtask A that asks to distinguish sexist and non-sexist texts. Additionally, we conduct experiments aimed at enhancing the interpretability of systems designed to detect online sexism. Our findings provide insights into the features and decision-making processes underlying our classifier system, thereby contributing to a broader effort to develop explainable AI models to detect online sexism.

In this paper, we describe M-FleNS, a multilingual flexible plug-and-play architecture designed to accommodate neural and symbolic modules, and initially instantiated with rule-based modules. We focus on using M-FleNS for the specific purpose of building new resources for Irish, a language currently under-represented in the NLP landscape. We present the general M-FleNS framework and how we use it to build an Irish Natural Language Generation system for verbalising part of the DBpedia ontology and building a multilayered dataset with rich linguistic annotations. Via automatic and human assessments of the output texts we show that with very limited resources we are able to create a system that reaches high levels of fluency and semantic accuracy, while having very low energy and memory requirements.

In this paper, we describe the submission of Dublin City University (DCU) and Trinity College Dublin (TCD) for the WebNLG 2023 shared task. We present a fully rule-based pipeline for generating Irish texts from DBpedia triple sets which comprises 4 components: triple lexicalisation, generation of noninflected Irish text, inflection generation, and post-processing.

Automated textual cyberbullying detection is known to be a challenging task. It is sometimes expected that messages associated with bullying will either be a) abusive, b) targeted at a specific individual or group, or c) have a negative sentiment. Transfer learning by fine-tuning pre-trained attention-based transformer language models (LMs) has achieved near state-of-the-art (SOA) precision in identifying textual fragments as being bullying-related or not. This study looks closely at two SOA LMs, BERT and HateBERT, fine-tuned on real-life cyberbullying datasets from multiple social networking platforms. We intend to determine whether these finely calibrated pre-trained LMs learn textual cyberbullying attributes or syntactical features in the text. The results of our comprehensive experiments show that despite the fact that attention weights are drawn more strongly to syntactical features of the text at every layer, attention weights cannot completely account for the decision-making of such attention-based transformers.

This paper presents baseline classification models for subjectivity detection, sentiment analysis, emotion analysis, sarcasm detection, and irony detection. All models are trained on user-generated content gathered from newswires and social networking services, in three different languages: English —a high-resourced language, Maltese —a low-resourced language, and Maltese-English —a code-switched language. Traditional supervised algorithms namely, Support Vector Machines, Naïve Bayes, Logistic Regression, Decision Trees, and Random Forest, are used to build a baseline for each classification task, namely subjectivity, sentiment polarity, emotion, sarcasm, and irony. Baseline models are established at a monolingual (English) level and at a code-switched level (Maltese-English). Results obtained from all the classification models are presented.

Cyberbullying is bullying perpetrated via the medium of modern communication technologies like social media networks and gaming platforms. Unfortunately, most existing datasets focusing on cyberbullying detection or classification are i) limited in number ii) usually targeted to one specific online social networking (OSN) platform, or iii) often contain low-quality annotations. In this study, we fine-tune and benchmark state of the art neural transformers for the binary classification of cyberbullying in social media texts, which is of high value to Natural Language Processing (NLP) researchers and computational social scientists. Furthermore, this work represents the first step toward building neural language models for cross OSN platform cyberbullying classification to make them as OSN platform agnostic as possible.

This paper presents multidimensional Social Opinion Mining on user-generated content gathered from newswires and social networking services in three different languages: English —a high-resourced language, Maltese —a low-resourced language, and Maltese-English —a code-switched language. Multiple fine-tuned neural classification language models which cater for the i) English, Maltese and Maltese-English languages as well as ii) five different social opinion dimensions, namely subjectivity, sentiment polarity, emotion, irony and sarcasm, are presented. Results per classification model for each social opinion dimension are discussed.

This study introduces an enriched version of the E2E dataset, one of the most popular language resources for data-to-text NLG. We extract intermediate representations for popular pipeline tasks such as discourse ordering, text structuring, lexicalization and referring expression generation, enabling researchers to rapidly develop and evaluate their data-to-text pipeline systems. The intermediate representations are extracted by aligning non-linguistic and text representations through a process called delexicalization, which consists in replacing input referring expressions to entities/attributes with placeholders. The enriched dataset is publicly available.

This paper reports results from a reproduction study in which we repeated the human evaluation of the PASS Dutch-language football report generation system (van der Lee et al., 2017). The work was carried out as part of the ReproGen Shared Task on Reproducibility of Human Evaluations in NLG, in Track A (Paper 1). We aimed to repeat the original study exactly, with the main difference that a different set of evaluators was used. We describe the study design, present the results from the original and the reproduction study, and then compare and analyse the differences between the two sets of results. For the two ‘headline’ results of average Fluency and Clarity, we find that in both studies, the system was rated more highly for Clarity than for Fluency, and Clarity had higher standard deviation. Clarity and Fluency ratings were higher, and their standard deviations lower, in the reproduction study than in the original study by substantial margins. Clarity had a higher degree of reproducibility than Fluency, as measured by the coefficient of variation. Data and code are publicly available.

The purpose of this paper is to present a prospective and interdisciplinary research project seeking to ontologize knowledge of the domain of Outsider Art, that is, the art created outside the boundaries of official culture. The goal is to combine ontology engineering methodologies to develop a knowledge base which i) examines the relation between social exclusion and cultural productions, ii) standardizes the terminology of Outsider Art and iii) enables semantic interoperability between cultural metadata relevant to Outsider Art. The Outsider Art ontology will integrate some existing ontologies and terminologies, such as the CIDOC - Conceptual Reference Model (CRM), the Art & Architecture Thesaurus and the Getty Union List of Artist Names, among other resources. Natural Language Processing and Machine Learning techniques will be fundamental instruments for knowledge acquisition and elicitation. NLP techniques will be used to annotate bibliographies of relevant outsider artists and descriptions of outsider artworks with linguistic information. Machine Learning techniques will be leveraged to acquire knowledge from linguistic features embedded in both types of texts.

The aim of this position paper is to establish an initial approach to the automatic classification of digital images about the Outsider Art style of painting. Specifically, we explore whether is it possible to classify non-traditional artistic styles by using the same features that are used for classifying traditional styles? Our research question is motivated by two facts. First, art historians state that non-traditional styles are influenced by factors “outside” of the world of art. Second, some studies have shown that several artistic styles confound certain classification techniques. Following current approaches to style prediction, this paper utilises Deep Learning methods to encode image features. Our preliminary experiments have provided motivation to think that, as is the case with traditional styles, Outsider Art can be computationally modelled with objective means by using training datasets and CNN models. Nevertheless, our results are not conclusive due to the lack of a large available dataset on Outsider Art. Therefore, at the end of the paper, we have mapped future lines of action, which include the compilation of a large dataset of Outsider Art images and the creation of an ontology of Outsider Art.

We present a gold standard of annotated social opinion for the Malta Government Budget 2018. It consists of over 500 online posts in English and/or the Maltese less-resourced language, gathered from social media platforms, specifically, social networking services and newswires, which have been annotated with information about opinions expressed by the general public and other entities, in terms of sentiment polarity, emotion, sarcasm/irony, and negation. This dataset is a resource for opinion mining based on social data, within the context of politics. It is the first opinion annotated social dataset from Malta, which has very limited language resources available.

FinSentiA: Sentiment Analysis in English Financial Microblogs The objective of this paper is to report on the building of a Sentiment Analysis (SA) system dedicated to financial microblogs in English. The purpose of our work is to build a financial classifier that predicts the sentiment of stock investors in microblog platforms such as StockTwits and Twitter. Our contribution shows that it is possible to conduct such tasks in order to provide fine grained SA of financial microblogs. We extracted financial entities with relevant contexts and assigned scores on a continuous scale by adopting a deep learning method for the classification.

This paper focuses on aspect extraction which is a sub-task of Aspect-based Sentiment Analysis. The goal is to report an extraction method of financial aspects in microblog messages. Our approach uses a stock-investment taxonomy for the identification of explicit and implicit aspects. We compare supervised and unsupervised methods to assign predefined categories at message level. Results on 7 aspect classes show 0.71 accuracy, while the 32 class classification gives 0.82 accuracy for messages containing explicit aspects and 0.35 for implicit aspects.

This paper discusses the “Fine-Grained Sentiment Analysis on Financial Microblogs and News” task as part of SemEval-2017, specifically under the “Detecting sentiment, humour, and truth” theme. This task contains two tracks, where the first one concerns Microblog messages and the second one covers News Statements and Headlines. The main goal behind both tracks was to predict the sentiment score for each of the mentioned companies/stocks. The sentiment scores for each text instance adopted floating point values in the range of -1 (very negative/bearish) to 1 (very positive/bullish), with 0 designating neutral sentiment. This task attracted a total of 32 participants, with 25 participating in Track 1 and 29 in Track 2.

The identification of semantic relations between terms within texts is a fundamental task in Natural Language Processing which can support applications requiring a lightweight semantic interpretation model. Currently, semantic relation classification concentrates on relations which are evaluated over open-domain data. This work provides a critique on the set of abstract relations used for semantic relation classification with regard to their ability to express relationships between terms which are found in a domain-specific corpora. Based on this analysis, this work proposes an alternative semantic relation model based on reusing and extending the set of abstract relations present in the DOLCE ontology. The resulting set of relations is well grounded, allows to capture a wide range of relations and could thus be used as a foundation for automatic classification of semantic relations.

Email can be considered as a virtual working environment in which users are constantly struggling to manage the vast amount of exchanged data. Although most of this data belongs to well-defined workflows, these are implicit and largely unsupported by existing email clients. Semanta provides this support by enabling Semantic Email ― email enhanced with machine-processable metadata about specific types of email Action Items (e.g. Task Assignment, Meeting Proposal). In the larger picture, these items form part of ad-hoc workflows (e.g. Task Delegation, Meeting Scheduling). Semanta is faced with a knowledge-acquisition bottleneck, as users cannot be expected to annotate each action item, and their automatic recognition proves difficult. This paper focuses on applying computationally treatable aspects of speech act theory for the classification of email action items. A rule-based classification model is employed, based on the presence or form of a number of linguistic features. The technology’s evaluation suggests that whereas full automation is not feasible, the results are good enough to be presented as suggestions for the user to review. In addition the rule-based system will bootstrap a machine learning system that is currently in development, to generate the initial training sets which are then improved through the user’s reviewing.

Although the Semantic web is steadily gaining in popularity, it remains a mystery to a large percentage of Internet users. This can be attributed to the complexity of the technologies that form its core. Creating intuitive interfaces which completely abstract the technologies underneath, is one way to solve this problem. A contrasting approach is to ease the user into understanding the technologies. We propose a solution which anchors on using controlled languages as interfaces to semantic web applications. This paper describes one such approach for the domain of meeting minutes, status reports and other project specific documents. A controlled language is developed along with an ontology to handle semi-automatic knowledge extraction. The contributions of this paper include an ontology designed for the domain of meeting minutes and status reports, and a controlled language grammar tailored for the above domain to perform the semi-automatic knowledge acquisition and generate RDF triples. This paper also describes two grammar prototypes, which were developed and evaluated prior to the development of the final grammar, as well as the Link grammar, which was the grammar formalism of choice.

The lack of structure in the content of email messages makes it very hard for data channelled between the sender and the recipient to be correctly interpreted and acted upon. As a result, the purposes of messages frequently end up not being fulfilled, prompting prolonged communication and stalling the disconnected workflow that is characteristic of email. This problem could be partially solved by extending the current email model to support light-weight semantics pertaining to the intents of the sender and the expectations from the recipient(s), thus leaving no room for ambiguity. Semantically-aware email clients will then be able to support the user with the workflow of email-generated tasks. In line with this thinking, we present the sMail Conceptual Framework. At its core, this framework has an Email Speech Act Model. Given this model, email content can be categorized into a set of speech acts, each carrying specific expectations. In this paper we present and discuss the methodology and results of this model?s statistical evaluation. By performing the same evaluation on another existing model, we demonstrate our model?s higher sophistication. After careful observations, we perform changes to the model and subsequently accommodate the changes in the revised sMail Conceptual Framework.

The identification of class instances within unstructured text for either the purposes of Ontology population or semantic annotation are usually limited to term mentions of Proper Noun and Personal Noun or fixed Key Phrases within Text Analytics or Ontology based Information Extraction(OBIE) applications. These systems do not generalize to cope with compound nominal classes of multi word expressions. Computational Linguistics’ approaches involving deep analysis tend to suffer from idiomaticity and overgeneration problems while the shallower “words with spaces” approach frequently employed in Information Extraction(IE) and Industrial Text Analytics systems lacks flexibility and is prone to lexical proliferation. We outline a representation for encoding light linguistic features of Compound Nominal term mentions of Concepts within an Ontology as well as a lightweight semantic annotator which complies the above linguistic information into efficient Dictionary formats to drive large scale identification and semantic annotation of the aforementioned concepts.