Gerasimos Spanakis

Thales Bertaglia, Haoyang Gui, Catalina Goanta et al.

Public consultations generate large volumes of data in the form of stakeholder submissions that are practically unfeasible to analyse manually. We present an end-to-end LLM-based pipeline and interactive dashboard for structured topic extraction from regulatory consultation submissions, demonstrated on the European Commission's Digital Fairness Act (DFA) public call for evidence as a case study. The system processes raw PDF attachments and web-form responses, extracts topic annotations, and grounds every extraction in a verbatim quote from the source text. Applied to 4,322 DFA submissions, the pipeline produced 15,368 topic annotations supported by 20,951 verbatim evidence quotes. Three principles govern the proposed design: verbatim grounding, full traceability, and transparency by design. The dashboard exposes the full extraction dataset through five analytical views, from dataset-level topic overviews to individual paragraph drill-downs, with every result traceable to its source. Beyond the predefined DFA topic categories, the pipeline generated certain stakeholder concerns, such as Age Verification, Payment Processor Censorship, and Digital Ownership, that a fixed-taxonomy approach would have missed. The pipeline is domain-generic; adapting it to a new consultation requires only a prompt update and a new dataset. A live demo is available at https://dfa-dashboard.thalesbertaglia.com/. The code and processed data are publicly available at https://github.com/thalesbertaglia/dfa-dashboard.

Catalina Goanta, Savvas Zannettou, Rishabh Kaushal et al.

To facilitate accountability and transparency, the Digital Services Act (DSA) sets up a process through which Very Large Online Platforms (VLOPs) need to grant vetted researchers access to their internal data (Article 40(4)). Operationalising such access is challenging for at least two reasons. First, data access is only available for research on systemic risks affecting European citizens, a concept with high levels of legal uncertainty. Second, data access suffers from an inherent standoff problem. Researchers need to request specific data but are not in a position to know all internal data processed by VLOPs, who, in turn, expect data specificity for potential access. In light of these limitations, data access under the DSA remains a mystery. To contribute to the discussion of how Article 40 can be interpreted and applied, we provide a concrete illustration of what data access can look like in a real-world systemic risk case study. We focus on the 2024 Romanian presidential election interference incident, the first event of its kind to trigger systemic risk investigations by the European Commission. During the elections, one candidate is said to have benefited from TikTok algorithmic amplification through a complex dis- and misinformation campaign. By analysing this incident, we can comprehend election-related systemic risk to explore practical research tasks and compare necessary data with available TikTok data. In particular, we make two contributions: (i) we combine insights from law, computer science and platform governance to shed light on the complexities of studying systemic risks in the context of election interference, focusing on two relevant factors: platform manipulation and hidden advertising; and (ii) we provide practical insights into various categories of available data for the study of TikTok, based on platform documentation, data donations and the Research API.

Mandani Ntekouli, Gerasimos Spanakis, Lourens Waldorp et al.

Previous research on EMA data of mental disorders was mainly focused on multivariate regression-based approaches modeling each individual separately. This paper goes a step further towards exploring the use of non-linear interpretable machine learning (ML) models in classification problems. ML models can enhance the ability to accurately predict the occurrence of different behaviors by recognizing complicated patterns between variables in data. To evaluate this, the performance of various ensembles of trees are compared to linear models using imbalanced synthetic and real-world datasets. After examining the distributions of AUC scores in all cases, non-linear models appear to be superior to baseline linear models. Moreover, apart from personalized approaches, group-level prediction models are also likely to offer an enhanced performance. According to this, two different nomothetic approaches to integrate data of more than one individuals are examined, one using directly all data during training and one based on knowledge distillation. Interestingly, it is observed that in one of the two real-world datasets, knowledge distillation method achieves improved AUC scores (mean relative change of +17\% compared to personalized) showing how it can benefit EMA data classification and performance.

Antoine Louis, Gijs van Dijck, Gerasimos Spanakis

Many individuals are likely to face a legal dispute at some point in their lives, but their lack of understanding of how to navigate these complex issues often renders them vulnerable. The advancement of natural language processing opens new avenues for bridging this legal literacy gap through the development of automated legal aid systems. However, existing legal question answering (LQA) approaches often suffer from a narrow scope, being either confined to specific legal domains or limited to brief, uninformative responses. In this work, we propose an end-to-end methodology designed to generate long-form answers to any statutory law questions, utilizing a "retrieve-then-read" pipeline. To support this approach, we introduce and release the Long-form Legal Question Answering (LLeQA) dataset, comprising 1,868 expert-annotated legal questions in the French language, complete with detailed answers rooted in pertinent legal provisions. Our experimental results demonstrate promising performance on automatic evaluation metrics, but a qualitative analysis uncovers areas for refinement. As one of the only comprehensive, expert-annotated long-form LQA dataset, LLeQA has the potential to not only accelerate research towards resolving a significant real-world issue, but also act as a rigorous benchmark for evaluating NLP models in specialized domains. We publicly release our code, data, and models.

Thales Bertaglia, Stefan Huber, Catalina Goanta et al.

Regulatory bodies worldwide are intensifying their efforts to ensure transparency in influencer marketing on social media through instruments like the Unfair Commercial Practices Directive (UCPD) in the European Union, or Section 5 of the Federal Trade Commission Act. Yet enforcing these obligations has proven to be highly problematic due to the sheer scale of the influencer market. The task of automatically detecting sponsored content aims to enable the monitoring and enforcement of such regulations at scale. Current research in this field primarily frames this problem as a machine learning task, focusing on developing models that achieve high classification performance in detecting ads. These machine learning tasks rely on human data annotation to provide ground truth information. However, agreement between annotators is often low, leading to inconsistent labels that hinder the reliability of models. To improve annotation accuracy and, thus, the detection of sponsored content, we propose using chatGPT to augment the annotation process with phrases identified as relevant features and brief explanations. Our experiments show that this approach consistently improves inter-annotator agreement and annotation accuracy. Additionally, our survey of user experience in the annotation task indicates that the explanations improve the annotators' confidence and streamline the process. Our proposed methods can ultimately lead to more transparency and alignment with regulatory requirements in sponsored content detection.

Mandani Ntekouli, Gerasimos Spanakis, Lourens Waldorp et al.

In the field of psychopathology, Ecological Momentary Assessment (EMA) methodological advancements have offered new opportunities to collect time-intensive, repeated and intra-individual measurements. This way, a large amount of data has become available, providing the means for further exploring mental disorders. Consequently, advanced machine learning (ML) methods are needed to understand data characteristics and uncover hidden and meaningful relationships regarding the underlying complex psychological processes. Among other uses, ML facilitates the identification of similar patterns in data of different individuals through clustering. This paper focuses on clustering multivariate time-series (MTS) data of individuals into several groups. Since clustering is an unsupervised problem, it is challenging to assess whether the resulting grouping is successful. Thus, we investigate different clustering methods based on different distance measures and assess them for the stability and quality of the derived clusters. These clustering steps are illustrated on a real-world EMA dataset, including 33 individuals and 15 variables. Through evaluation, the results of kernel-based clustering methods appear promising to identify meaningful groups in the data. So, efficient representations of EMA data play an important role in clustering.

Mandani Ntekouli, Gerasimos Spanakis, Lourens Waldorp et al.

Through Ecological Momentary Assessment (EMA) studies, a number of time-series data is collected across multiple individuals, continuously monitoring various items of emotional behavior. Such complex data is commonly analyzed in an individual level, using personalized models. However, it is believed that additional information of similar individuals is likely to enhance these models leading to better individuals' description. Thus, clustering is investigated with an aim to group together the most similar individuals, and subsequently use this information in group-based models in order to improve individuals' predictive performance. More specifically, two model-based clustering approaches are examined, where the first is using model-extracted parameters of personalized models, whereas the second is optimized on the model-based forecasting performance. Both methods are then analyzed using intrinsic clustering evaluation measures (e.g. Silhouette coefficients) as well as the performance of a downstream forecasting scheme, where each forecasting group-model is devoted to describe all individuals belonging to one cluster. Among these, clustering based on performance shows the best results, in terms of all examined evaluation measures. As another level of evaluation, those group-models' performance is compared to three baseline scenarios, the personalized, the all-in-one group and the random group-based concept. According to this comparison, the superiority of clustering-based methods is again confirmed, indicating that the utilization of group-based information could be effectively enhance the overall performance of all individuals' data.

Catalina Goanta, Nikolaos Aletras, Ilias Chalkidis et al.

The scientific innovation in Natural Language Processing (NLP) and more broadly in artificial intelligence (AI) is at its fastest pace to date. As large language models (LLMs) unleash a new era of automation, important debates emerge regarding the benefits and risks of their development, deployment and use. Currently, these debates have been dominated by often polarized narratives mainly led by the AI Safety and AI Ethics movements. This polarization, often amplified by social media, is swaying political agendas on AI regulation and governance and posing issues of regulatory capture. Capture occurs when the regulator advances the interests of the industry it is supposed to regulate, or of special interest groups rather than pursuing the general public interest. Meanwhile in NLP research, attention has been increasingly paid to the discussion of regulating risks and harms. This often happens without systematic methodologies or sufficient rooting in the disciplines that inspire an extended scope of NLP research, jeopardizing the scientific integrity of these endeavors. Regulation studies are a rich source of knowledge on how to systematically deal with risk and uncertainty, as well as with scientific evidence, to evaluate and compare regulatory options. This resource has largely remained untapped so far. In this paper, we argue how NLP research on these topics can benefit from proximity to regulatory studies and adjacent fields. We do so by discussing basic tenets of regulation, and risk and uncertainty, and by highlighting the shortcomings of current NLP discussions dealing with risk assessment. Finally, we advocate for the development of a new multidisciplinary research space on regulation and NLP (RegNLP), focused on connecting scientific knowledge to regulatory processes based on systematic methodologies.

Antoine Louis, Gijs van Dijck, Gerasimos Spanakis

Statutory article retrieval (SAR), the task of retrieving statute law articles relevant to a legal question, is a promising application of legal text processing. In particular, high-quality SAR systems can improve the work efficiency of legal professionals and provide basic legal assistance to citizens in need at no cost. Unlike traditional ad-hoc information retrieval, where each document is considered a complete source of information, SAR deals with texts whose full sense depends on complementary information from the topological organization of statute law. While existing works ignore these domain-specific dependencies, we propose a novel graph-augmented dense statute retriever (G-DSR) model that incorporates the structure of legislation via a graph neural network to improve dense retrieval performance. Experimental results show that our approach outperforms strong retrieval baselines on a real-world expert-annotated SAR dataset.

Haoyang Gui, Thales Bertaglia, Catalina Goanta et al.

Content monetization on social media fuels a growing influencer economy. Influencer marketing remains largely undisclosed or inappropriately disclosed on social media. Non-disclosure issues have become a priority for national and supranational authorities worldwide, who are starting to impose increasingly harsher sanctions on them. This paper proposes a transparent methodology for measuring whether and how influencers comply with disclosures based on legal standards. We introduce a novel distinction between disclosures that are legally sufficient (green) and legally insufficient (yellow). We apply this methodology to an original dataset reflecting the content of 150 Dutch influencers publicly registered with the Dutch Media Authority based on recently introduced registration obligations. The dataset consists of 292,315 posts and is multi-language (English and Dutch) and cross-platform (Instagram, YouTube and TikTok). We find that influencer marketing remains generally underdisclosed on social media, and that bigger influencers are not necessarily more compliant with disclosure standards.

Vageesh Saxena, Benjamin Bashpole, Gijs Van Dijck et al.

Human trafficking (HT) is a pervasive global issue affecting vulnerable individuals, violating their fundamental human rights. Investigations reveal that a significant number of HT cases are associated with online advertisements (ads), particularly in escort markets. Consequently, identifying and connecting HT vendors has become increasingly challenging for Law Enforcement Agencies (LEAs). To address this issue, we introduce IDTraffickers, an extensive dataset consisting of 87,595 text ads and 5,244 vendor labels to enable the verification and identification of potential HT vendors on online escort markets. To establish a benchmark for authorship identification, we train a DeCLUTR-small model, achieving a macro-F1 score of 0.8656 in a closed-set classification environment. Next, we leverage the style representations extracted from the trained classifier to conduct authorship verification, resulting in a mean r-precision score of 0.8852 in an open-set ranking environment. Finally, to encourage further research and ensure responsible data sharing, we plan to release IDTraffickers for the authorship attribution task to researchers under specific conditions, considering the sensitive nature of the data. We believe that the availability of our dataset and benchmarks will empower future researchers to utilize our findings, thereby facilitating the effective linkage of escort ads and the development of more robust approaches for identifying HT indicators.

Antoine Louis, Gijs van Dijck, Gerasimos Spanakis

Hybrid search has emerged as an effective strategy to offset the limitations of different matching paradigms, especially in out-of-domain contexts where notable improvements in retrieval quality have been observed. However, existing research predominantly focuses on a limited set of retrieval methods, evaluated in pairs on domain-general datasets exclusively in English. In this work, we study the efficacy of hybrid search across a variety of prominent retrieval models within the unexplored field of law in the French language, assessing both zero-shot and in-domain scenarios. Our findings reveal that in a zero-shot context, fusing different domain-general models consistently enhances performance compared to using a standalone model, regardless of the fusion method. Surprisingly, when models are trained in-domain, we find that fusion generally diminishes performance relative to using the best single system, unless fusing scores with carefully tuned weights. These novel insights, among others, expand the applicability of prior findings across a new field and language, and contribute to a deeper understanding of hybrid search in non-English specialized domains.

Abderrahmane Issam, Yusuf Can Semerci, Jan Scholtes et al.

Simultaneous machine translation aims at solving the task of real-time translation by starting to translate before consuming the full input, which poses challenges in terms of balancing quality and latency of the translation. The wait-$k$ policy offers a solution by starting to translate after consuming $k$ words, where the choice of the number $k$ directly affects the latency and quality. In applications where we seek to keep the choice over latency and quality at inference, the wait-$k$ policy obliges us to train more than one model. In this paper, we address the challenge of building one model that can fulfil multiple latency levels and we achieve this by introducing lightweight adapter modules into the decoder. The adapters are trained to be specialized for different wait-$k$ values and compared to other techniques they offer more flexibility to allow for reaping the benefits of parameter sharing and minimizing interference. Additionally, we show that by combining with an adaptive strategy, we can further improve the results. Experiments on two language directions show that our method outperforms or competes with other strong baselines on most latency values.

Gianluca Vico, Gerasimos Spanakis

Etruscan is an ancient language spoken in Italy from the 7th century BC to the 1st century AD. There are no native speakers of the language at the present day, and its resources are scarce, as there exist only around 12,000 known inscriptions. To the best of our knowledge, there are no publicly available Etruscan corpora for natural language processing. Therefore, we propose a dataset for machine translation from Etruscan to English, which contains 2891 translated examples from existing academic sources. Some examples are extracted manually, while others are acquired in an automatic way. Along with the dataset, we benchmark different machine translation models observing that it is possible to achieve a BLEU score of 10.1 with a small transformer model. Releasing the dataset can help enable future research on this language, similar languages or other languages with scarce resources.

Justus-Jonas Erker, Stefan Schaffer, Gerasimos Spanakis

Inspired by the curvature of space-time (Einstein, 1921), we introduce Curved Contrastive Learning (CCL), a novel representation learning technique for learning the relative turn distance between utterance pairs in multi-turn dialogues. The resulting bi-encoder models can guide transformers as a response ranking model towards a goal in a zero-shot fashion by projecting the goal utterance and the corresponding reply candidates into a latent space. Here the cosine similarity indicates the distance/reachability of a candidate utterance toward the corresponding goal. Furthermore, we explore how these forward-entailing language representations can be utilized for assessing the likelihood of sequences by the entailment strength i.e. through the cosine similarity of its individual members (encoded separately) as an emergent property in the curved space. These non-local properties allow us to imagine the likelihood of future patterns in dialogues, specifically by ordering/identifying future goal utterances that are multiple turns away, given a dialogue context. As part of our analysis, we investigate characteristics that make conversations (un)plannable and find strong evidence of planning capability over multiple turns (in 61.56% over 3 turns) in conversations from the DailyDialog (Li et al., 2017) dataset. Finally, we show how we achieve higher efficiency in sequence modeling tasks compared to previous work thanks to our relativistic approach, where only the last utterance needs to be encoded and computed during inference.

Abderrahmane Issam, Yusuf Can Semerci, Jan Scholtes et al.

Despite the centrality of crosslinguistic influence (CLI) to bilingualism research, human studies often yield conflicting results due to inherent experimental variance. We address these inconsistencies by using language models (LMs) as controlled statistical learners to systematically simulate CLI and isolate its underlying drivers. Specifically, we study the effect of varying the L1 language dominance and the L2 language proficiency, which we manipulate by controlling the L2 age of exposure -- defined as the training step at which the L2 is introduced. Furthermore, we investigate the impact of pretraining on L1 languages with varying syntactic distance from the L2. Using cross-linguistic priming, we analyze how activating L1 structures impacts L2 processing. Our results align with evidence from psycholinguistic studies, confirming that language dominance and proficiency are strong predictors of CLI. We further find that while priming of grammatical structures is bidirectional, the priming of ungrammatical structures is sensitive to language dominance. Finally, we provide mechanistic evidence of CLI in LMs, demonstrating that the L1 is co-activated during L2 processing and directly influences the neural circuitry recruited for the L2. More broadly, our work demonstrates that LMs can serve as a computational framework to inform theories of human CLI.

Antoine Louis, Vageesh Saxena, Gijs van Dijck et al.

State-of-the-art neural retrievers predominantly focus on high-resource languages like English, which impedes their adoption in retrieval scenarios involving other languages. Current approaches circumvent the lack of high-quality labeled data in non-English languages by leveraging multilingual pretrained language models capable of cross-lingual transfer. However, these models require substantial task-specific fine-tuning across multiple languages, often perform poorly in languages with minimal representation in the pretraining corpus, and struggle to incorporate new languages after the pretraining phase. In this work, we present a novel modular dense retrieval model that learns from the rich data of a single high-resource language and effectively zero-shot transfers to a wide array of languages, thereby eliminating the need for language-specific labeled data. Our model, ColBERT-XM, demonstrates competitive performance against existing state-of-the-art multilingual retrievers trained on more extensive datasets in various languages. Further analysis reveals that our modular approach is highly data-efficient, effectively adapts to out-of-distribution data, and significantly reduces energy consumption and carbon emissions. By demonstrating its proficiency in zero-shot scenarios, ColBERT-XM marks a shift towards more sustainable and inclusive retrieval systems, enabling effective information accessibility in numerous languages. We publicly release our code and models for the community.

Lucas G. Uberti-Bona Marin, Bram Rijsbosch, Kristof Meding et al.

Technical and legal debates frequently suggest that "accuracy" is an objective, measurable, and purely technical property. We challenge this view, showing that evaluating AI performance fundamentally depends on context-dependent normative decisions. These techno-normative choices are crucial for rigorous AI deployment, as they determine which errors are prioritised, how risks are distributed, and how trade-offs between competing objectives are resolved. This paper provides a legal-technical analysis of the choices that shape how accuracy is defined, measured, and assessed, using the 2024 European Union AI Act -- which mandates an "appropriate level of accuracy" for high-risk systems -- as a primary case study. We identify and analyse four choices central to any robust performance evaluation: (1) selecting metrics, (2) balancing multiple metrics, (3) measuring metrics against representative data, and (4) determining acceptance thresholds. For each choice, we study its relationship to the AI Act's accuracy requirement and associated documentation obligations, show how its technical implementation embeds implicit or explicit assumptions about acceptable risks, errors, and trade-offs, and discuss the implications for the practical implementation of the AI Act by examples and related technical standards. By making the techno-normative dimensions of accuracy explicit, this paper contributes to broader interdisciplinary debates on AI governance and regulation, and offers specific guidance for regulators, auditors, and developers tasked with translating (legal) safety requirements into technical practice.

Justus-Jonas Erker, Florian Mai, Nils Reimers et al. · huggingface

Search-based dialog models typically re-encode the dialog history at every turn, incurring high cost. Curved Contrastive Learning, a representation learning method that encodes relative distances between utterances into the embedding space via a bi-encoder, has recently shown promising results for dialog modeling at far superior efficiency. While high efficiency is achieved through independently encoding utterances, this ignores the importance of contextualization. To overcome this issue, this study introduces triple-encoders, which efficiently compute distributed utterance mixtures from these independently encoded utterances through a novel hebbian inspired co-occurrence learning objective in a self-organizing manner, without using any weights, i.e., merely through local interactions. Empirically, we find that triple-encoders lead to a substantial improvement over bi-encoders, and even to better zero-shot generalization than single-vector representation models without requiring re-encoding. Our code (https://github.com/UKPLab/acl2024-triple-encoders) and model (https://huggingface.co/UKPLab/triple-encoders-dailydialog) are publicly available.

Dominika Woszczyk, Gerasimos Spanakis

Urbanism is no longer planned on paper thanks to powerful models and 3D simulation platforms. However, current work is not open to the public and lacks an optimisation agent that could help in decision making. This paper describes the creation of an open-source simulation based on an existing Dutch liveability score with a built-in AI module. Features are selected using feature engineering and Random Forests. Then, a modified scoring function is built based on the former liveability classes. The score is predicted using Random Forest for regression and achieved a recall of 0.83 with 10-fold cross-validation. Afterwards, Exploratory Factor Analysis is applied to select the actions present in the model. The resulting indicators are divided into 5 groups, and 12 actions are generated. The performance of four optimisation algorithms is compared, namely NSGA-II, PAES, SPEA2 and eps-MOEA, on three established criteria of quality: cardinality, the spread of the solutions, spacing, and the resulting score and number of turns. Although all four algorithms show different strengths, eps-MOEA is selected to be the most suitable for this problem. Ultimately, the simulation incorporates the model and the selected AI module in a GUI written in the Kivy framework for Python. Tests performed on users show positive responses and encourage further initiatives towards joining technology and public applications.

Abderrahmane Issam, Yusuf Can Semerci, Jan Scholtes et al.

End-to-End Speech Translation (E2E-ST) has seen significant advancements, yet current models are primarily benchmarked on curated, "clean" datasets. This overlooks critical real-world challenges, such as morphological robustness to inflectional variations common in non-native or dialectal speech. In this work, we adapt a text-based adversarial attack targeting inflectional morphology to the speech domain and demonstrate that state-of-the-art E2E-ST models are highly vulnerable it. While adversarial training effectively mitigates such risks in text-based tasks, generating high-quality adversarial speech data remains computationally expensive and technically challenging. To address this, we propose Cross-Modal Robustness Transfer (CMRT), a framework that transfers adversarial robustness from the text modality to the speech modality. Our method eliminates the requirement for adversarial speech data during training. Extensive experiments across four language pairs demonstrate that CMRT improves adversarial robustness by an average of more than 3 BLEU points, establishing a new baseline for robust E2E-ST without the overhead of generating adversarial speech.

Ben Hagag, Liav Harpaz, Gil Semo et al.

This paper presents the results of the LegalLens Shared Task, focusing on detecting legal violations within text in the wild across two sub-tasks: LegalLens-NER for identifying legal violation entities and LegalLens-NLI for associating these violations with relevant legal contexts and affected individuals. Using an enhanced LegalLens dataset covering labor, privacy, and consumer protection domains, 38 teams participated in the task. Our analysis reveals that while a mix of approaches was used, the top-performing teams in both tasks consistently relied on fine-tuning pre-trained language models, outperforming legal-specific models and few-shot methods. The top-performing team achieved a 7.11% improvement in NER over the baseline, while NLI saw a more marginal improvement of 5.7%. Despite these gains, the complexity of legal texts leaves room for further advancements.

Abdulmuizz Khalak, Abderrahmane Issam, Gerasimos Spanakis

Arabic Language Models (LMs) are pretrained predominately on Modern Standard Arabic (MSA) and are expected to transfer to its dialects. While MSA as the standard written variety is commonly used in formal settings, people speak and write online in various dialects that are spread across the Arab region. This poses limitations for Arabic LMs, since its dialects vary in their similarity to MSA. In this work we study cross-lingual transfer of Arabic models using probing on 3 Natural Language Processing (NLP) Tasks, and representational similarity. Our results indicate that transfer is possible but disproportionate across dialects, which we find to be partially explained by their geographic proximity. Furthermore, we find evidence for negative interference in models trained to support all Arabic dialects. This questions their degree of similarity, and raises concerns for cross-lingual transfer in Arabic models.

Lucas-Andreï Thil, Mirela Popa, Gerasimos Spanakis

Recent advancements in language models have demonstrated remarkable improvements in various natural language processing (NLP) tasks such as web navigation. Supervised learning (SL) approaches have achieved impressive performance while utilizing significantly less training data compared to previous methods. However, these SL-based models fall short when compared to reinforcement learning (RL) approaches, which have shown superior results. In this paper, we propose a novel approach that combines SL and RL techniques over the MiniWoB benchmark to leverage the strengths of both methods. We also address a critical limitation in previous models' understanding of HTML content, revealing a tendency to memorize target elements rather than comprehend the underlying structure. To rectify this, we propose methods to enhance true understanding and present a new baseline of results. Our experiments demonstrate that our approach outperforms previous SL methods on certain tasks using less data and narrows the performance gap with RL models, achieving 43.58\% average accuracy in SL and 36.69\% when combined with a multimodal RL approach. This study sets a new direction for future web navigation and offers insights into the limitations and potential of language modeling for computer tasks.

Paweł Mąka, Yusuf Can Semerci, Jan Scholtes et al.

Context-aware Machine Translation aims to improve translations of sentences by incorporating surrounding sentences as context. Towards this task, two main architectures have been applied, namely single-encoder (based on concatenation) and multi-encoder models. In this study, we show that a special case of multi-encoder architecture, where the latent representation of the source sentence is cached and reused as the context in the next step, achieves higher accuracy on the contrastive datasets (where the models have to rank the correct translation among the provided sentences) and comparable BLEU and COMET scores as the single- and multi-encoder approaches. Furthermore, we investigate the application of Sequence Shortening to the cached representations. We test three pooling-based shortening techniques and introduce two novel methods - Latent Grouping and Latent Selecting, where the network learns to group tokens or selects the tokens to be cached as context. Our experiments show that the two methods achieve competitive BLEU and COMET scores and accuracies on the contrastive datasets to the other tested methods while potentially allowing for higher interpretability and reducing the growth of memory requirements with increased context size.

Haoyang Gui, Thales Bertaglia, Catalina Goanta et al.

Influencer marketing has become a crucial feature of digital marketing strategies. Despite its rapid growth and algorithmic relevance, the field of computational studies in influencer marketing remains fragmented, especially with limited systematic reviews covering the computational methodologies employed. This makes overarching scientific measurements in the influencer economy very scarce, to the detriment of interested stakeholders outside of platforms themselves, such as regulators, but also researchers from other fields. This paper aims to provide an overview of the state of the art of computational studies in influencer marketing by conducting a systematic literature review (SLR) based on the PRISMA model. The paper analyses 69 studies to identify key research themes, methodologies, and future directions in this research field. The review identifies four major research themes: Influencer identification and characterisation, Advertising strategies and engagement, Sponsored content analysis and discovery, and Fairness. Methodologically, the studies are categorised into machine learning-based techniques (e.g., classification, clustering) and non-machine-learning-based techniques (e.g., statistical analysis, network analysis). Key findings reveal a strong focus on optimising commercial outcomes, with limited attention to regulatory compliance and ethical considerations. The review highlights the need for more nuanced computational research that incorporates contextual factors such as language, platform, and industry type, as well as improved model explainability and dataset reproducibility. The paper concludes by proposing a multidisciplinary research agenda that emphasises the need for further links to regulation and compliance technology, finer granularity in analysis, and the development of standardised datasets.

Paweł Mąka, Yusuf Can Semerci, Jan Scholtes et al.

In this paper, we investigate the role of attention heads in Context-aware Machine Translation models for pronoun disambiguation in the English-to-German and English-to-French language directions. We analyze their influence by both observing and modifying the attention scores corresponding to the plausible relations that could impact a pronoun prediction. Our findings reveal that while some heads do attend the relations of interest, not all of them influence the models' ability to disambiguate pronouns. We show that certain heads are underutilized by the models, suggesting that model performance could be improved if only the heads would attend one of the relations more strongly. Furthermore, we fine-tune the most promising heads and observe the increase in pronoun disambiguation accuracy of up to 5 percentage points which demonstrates that the improvements in performance can be solidified into the models' parameters.

Haoyang Gui, Thales Bertaglia, Taylor Annabell et al.

The rise of influencer marketing has blurred boundaries between organic content and sponsored content, making the enforcement of legal rules relating to transparency challenging. Effective regulation requires applying legal knowledge with a clear purpose and reason, yet current detection methods of undisclosed sponsored content generally lack legal grounding or operate as opaque "black boxes". Using 1,143 Instagram posts, we compare gpt-5-nano and gemini-2.5-flash-lite under three prompting strategies with controlled levels of legal knowledge provided. Both models perform strongly in classifying content as sponsored or not (F1 up to 0.93), though performance drops by over 10 points on ambiguous cases. We further develop a taxonomy of reasoning errors, showing frequent citation omissions (28.57%), unclear references (20.71%), and hidden ads exhibiting the highest miscue rate (28.57%). While adding regulatory text to the prompt improves explanation quality, it does not consistently improve detection accuracy. The contribution of this paper is threefold. First, it makes a novel addition to regulatory compliance technology by providing a taxonomy of common errors in LLM-generated legal reasoning to evaluate whether automated moderation is not only accurate but also legally robust, thereby advancing the transparent detection of influencer marketing content. Second, it features an original dataset of LLM explanations annotated by two students who were trained in influencer marketing law. Third, it combines quantitative and qualitative evaluation strategies for LLM explanations and critically reflects on how these findings can support advertising regulatory bodies in automating moderation processes on a solid legal foundation.

Abderrahmane Issam, Yusuf Can Semerci, Jan Scholtes et al.

End-to-End Speech Translation (E2E-ST) is the task of translating source speech directly into target text bypassing the intermediate transcription step. The representation discrepancy between the speech and text modalities has motivated research on what is known as bridging the modality gap. State-of-the-art methods addressed this by aligning speech and text representations on the word or token level. Unfortunately, this requires an alignment tool that is not available for all languages. Although this issue has been addressed by aligning speech and text embeddings using nearest-neighbor similarity search, it does not lead to accurate alignments. In this work, we adapt Dynamic Time Warping (DTW) for aligning speech and text embeddings during training. Our experiments demonstrate the effectiveness of our method in bridging the modality gap in E2E-ST. Compared to previous work, our method produces more accurate alignments and achieves comparable E2E-ST results while being significantly faster. Furthermore, our method outperforms previous work in low resource settings on 5 out of 6 language directions.

Paweł Mąka, Yusuf Can Semerci, Jan Scholtes et al.

Achieving human-level translations requires leveraging context to ensure coherence and handle complex phenomena like pronoun disambiguation. Sparsity of contextually rich examples in the standard training data has been hypothesized as the reason for the difficulty of context utilization. In this work, we systematically validate this claim in both single- and multilingual settings by constructing training datasets with a controlled proportions of contextually relevant examples. We demonstrate a strong association between training data sparsity and model performance confirming sparsity as a key bottleneck. Importantly, we reveal that improvements in one contextual phenomenon do no generalize to others. While we observe some cross-lingual transfer, it is not significantly higher between languages within the same sub-family. Finally, we propose and empirically evaluate two training strategies designed to leverage the available data. These strategies improve context utilization, resulting in accuracy gains of up to 6 and 8 percentage points on the ctxPro evaluation in single- and multilingual settings respectively.

Lucas G. Uberti-Bona Marin, Bram Rijsbosch, Gerasimos Spanakis et al.

As Artificial Intelligence becomes increasingly central to corporate strategies, concerns over its risks are growing too. In response, regulators are pushing for greater transparency in how companies identify, report and mitigate AI-related risks. In the US, the Securities and Exchange Commission (SEC) repeatedly warned companies to provide their investors with more accurate disclosures of AI-related risks; recent enforcement and litigation against companies' misleading AI claims reinforce these warnings. In the EU, new laws - like the AI Act and Digital Services Act - introduced additional rules on AI risk reporting and mitigation. Given these developments, it is essential to examine if and how companies report AI-related risks to the public. This study presents the first large-scale systematic analysis of AI risk disclosures in SEC 10-K filings, which require public companies to report material risks to their company. We analyse over 30,000 filings from more than 7,000 companies over the past five years, combining quantitative and qualitative analysis. Our findings reveal a sharp increase in the companies that mention AI risk, up from 4% in 2020 to over 43% in the most recent 2024 filings. While legal and competitive AI risks are the most frequently mentioned, we also find growing attention to societal AI risks, such as cyberattacks, fraud, and technical limitations of AI systems. However, many disclosures remain generic or lack details on mitigation strategies, echoing concerns raised recently by the SEC about the quality of AI-related risk reporting. To support future research, we publicly release a web-based tool for easily extracting and analysing keyword-based disclosures across SEC filings.

Elza Strazda, Gerasimos Spanakis

Warning: This paper contains explicit statements of offensive stereotypes which might be upsetting. Language models are prone to exhibiting biases, further amplifying unfair and harmful stereotypes. Given the fast-growing popularity and wide application of these models, it is necessary to ensure safe and fair language models. As of recent considerable attention has been paid to measuring bias in language models, yet the majority of studies have focused only on English language. A Dutch version of the US-specific CrowS-Pairs dataset for measuring bias in Dutch language models is introduced. The resulting dataset consists of 1463 sentence pairs that cover bias in 9 categories, such as Sexual orientation, Gender and Disability. The sentence pairs are composed of contrasting sentences, where one of the sentences concerns disadvantaged groups and the other advantaged groups. Using the Dutch CrowS-Pairs dataset, we show that various language models, BERTje, RobBERT, multilingual BERT, GEITje and Mistral-7B exhibit substantial bias across the various bias categories. Using the English and French versions of the CrowS-Pairs dataset, bias was evaluated in English (BERT and RoBERTa) and French (FlauBERT and CamemBERT) language models, and it was shown that English models exhibit the most bias, whereas Dutch models the least amount of bias. Additionally, results also indicate that assigning a persona to a language model changes the level of bias it exhibits. These findings highlight the variability of bias across languages and contexts, suggesting that cultural and linguistic factors play a significant role in shaping model biases.

Abderrahmane Issam, Yusuf Can Semerci, Jan Scholtes et al.

Understanding robustness is essential for building reliable NLP systems. Unfortunately, in the context of machine translation, previous work mainly focused on documenting robustness failures or improving robustness. In contrast, we study robustness from a model representation perspective by looking at internal model representations of ungrammatical inputs and how they evolve through model layers. For this purpose, we perform Grammatical Error Detection (GED) probing and representational similarity analysis. Our findings indicate that the encoder first detects the grammatical error, then corrects it by moving its representation toward the correct form. To understand what contributes to this process, we turn to the attention mechanism where we identify what we term Robustness Heads. We find that Robustness Heads attend to interpretable linguistic units when responding to grammatical errors, and that when we fine-tune models for robustness, they tend to rely more on Robustness Heads for updating the ungrammatical word representation.

Vageesh Saxena, Benjamin Bashpole, Gijs Van Dijck et al.

Human trafficking (HT) remains a critical issue, with traffickers increasingly leveraging online escort advertisements (ads) to advertise victims anonymously. Existing detection methods, including Authorship Attribution (AA), often center on text-based analyses and neglect the multimodal nature of online escort ads, which typically pair text with images. To address this gap, we introduce MATCHED, a multimodal dataset of 27,619 unique text descriptions and 55,115 unique images collected from the Backpage escort platform across seven U.S. cities in four geographical regions. Our study extensively benchmarks text-only, vision-only, and multimodal baselines for vendor identification and verification tasks, employing multitask (joint) training objectives that achieve superior classification and retrieval performance on in-distribution and out-of-distribution (OOD) datasets. Integrating multimodal features further enhances this performance, capturing complementary patterns across text and images. While text remains the dominant modality, visual data adds stylistic cues that enrich model performance. Moreover, text-image alignment strategies like CLIP and BLIP2 struggle due to low semantic overlap and vague connections between the modalities of escort ads, with end-to-end multimodal training proving more robust. Our findings emphasize the potential of multimodal AA (MAA) to combat HT, providing LEAs with robust tools to link ads and disrupt trafficking networks.

Mandani Ntekouli, Gerasimos Spanakis, Lourens Waldorp et al.

In the field of psychopathology, Ecological Momentary Assessment (EMA) studies offer rich individual data on psychopathology-relevant variables (e.g., affect, behavior, etc) in real-time. EMA data is collected dynamically, represented as complex multivariate time series (MTS). Such information is crucial for a better understanding of mental disorders at the individual- and group-level. More specifically, clustering individuals in EMA data facilitates uncovering and studying the commonalities as well as variations of groups in the population. Nevertheless, since clustering is an unsupervised task and true EMA grouping is not commonly available, the evaluation of clustering is quite challenging. An important aspect of evaluation is clustering explainability. Thus, this paper proposes an attention-based interpretable framework to identify the important time-points and variables that play primary roles in distinguishing between clusters. A key part of this study is to examine ways to analyze, summarize, and interpret the attention weights as well as evaluate the patterns underlying the important segments of the data that differentiate across clusters. To evaluate the proposed approach, an EMA dataset of 187 individuals grouped in 3 clusters is used for analyzing the derived attention-based importance attributes. More specifically, this analysis provides the distinct characteristics at the cluster-, feature- and individual level. Such clustering explanations could be beneficial for generalizing existing concepts of mental disorders, discovering new insights, and even enhancing our knowledge at an individual level.

Mandani Ntekouli, Gerasimos Spanakis, Lourens Waldorp et al.

In the evolving field of psychopathology, the accurate assessment and forecasting of data derived from Ecological Momentary Assessment (EMA) is crucial. EMA offers contextually-rich psychopathological measurements over time, that practically lead to Multivariate Time Series (MTS) data. Thus, many challenges arise in analysis from the temporal complexities inherent in emotional, behavioral, and contextual EMA data as well as their inter-dependencies. To address both of these aspects, this research investigates the performance of Recurrent and Temporal Graph Neural Networks (GNNs). Overall, GNNs, by incorporating additional information from graphs reflecting the inner relationships between the variables, notably enhance the results by decreasing the Mean Squared Error (MSE) to 0.84 compared to the baseline LSTM model at 1.02. Therefore, the effect of constructing graphs with different characteristics on GNN performance is also explored. Additionally, GNN-learned graphs, which are dynamically refined during the training process, were evaluated. Using such graphs showed a similarly good performance. Thus, graph learning proved also promising for other GNN methods, potentially refining the pre-defined graphs.

Vageesh Saxena, Nils Rethmeier, Gijs Van Dijck et al.

The anonymity on the Darknet allows vendors to stay undetected by using multiple vendor aliases or frequently migrating between markets. Consequently, illegal markets and their connections are challenging to uncover on the Darknet. To identify relationships between illegal markets and their vendors, we propose VendorLink, an NLP-based approach that examines writing patterns to verify, identify, and link unique vendor accounts across text advertisements (ads) on seven public Darknet markets. In contrast to existing literature, VendorLink utilizes the strength of supervised pre-training to perform closed-set vendor verification, open-set vendor identification, and low-resource market adaption tasks. Through VendorLink, we uncover (i) 15 migrants and 71 potential aliases in the Alphabay-Dreams-Silk dataset, (ii) 17 migrants and 3 potential aliases in the Valhalla-Berlusconi dataset, and (iii) 75 migrants and 10 potential aliases in the Traderoute-Agora dataset. Altogether, our approach can help Law Enforcement Agencies (LEA) make more informed decisions by verifying and identifying migrating vendors and their potential aliases on existing and Low-Resource (LR) emerging Darknet markets.

Antoine Louis, Gerasimos Spanakis

Statutory article retrieval is the task of automatically retrieving law articles relevant to a legal question. While recent advances in natural language processing have sparked considerable interest in many legal tasks, statutory article retrieval remains primarily untouched due to the scarcity of large-scale and high-quality annotated datasets. To address this bottleneck, we introduce the Belgian Statutory Article Retrieval Dataset (BSARD), which consists of 1,100+ French native legal questions labeled by experienced jurists with relevant articles from a corpus of 22,600+ Belgian law articles. Using BSARD, we benchmark several state-of-the-art retrieval approaches, including lexical and dense architectures, both in zero-shot and supervised setups. We find that fine-tuned dense retrieval models significantly outperform other systems. Our best performing baseline achieves 74.8% R@100, which is promising for the feasibility of the task and indicates there is still room for improvement. By the specificity of the domain and addressed task, BSARD presents a unique challenge problem for future research on legal information retrieval. Our dataset and source code are publicly available.

Błażej Dolicki, Gerasimos Spanakis

There is an increasing amount of evidence that in cases with little or no data in a target language, training on a different language can yield surprisingly good results. However, currently there are no established guidelines for choosing the training (source) language. In attempt to solve this issue we thoroughly analyze a state-of-the-art multilingual model and try to determine what impacts good transfer between languages. As opposed to the majority of multilingual NLP literature, we don't only train on English, but on a group of almost 30 languages. We show that looking at particular syntactic features is 2-4 times more helpful in predicting the performance than an aggregated syntactic similarity. We find out that the importance of syntactic features strongly differs depending on the downstream task - no single feature is a good performance predictor for all NLP tasks. As a result, one should not expect that for a target language $L_1$ there is a single language $L_2$ that is the best choice for any NLP task (for instance, for Bulgarian, the best source language is French on POS tagging, Russian on NER and Thai on NLI). We discuss the most important linguistic features affecting the transfer quality using statistical and machine learning methods.

Adam Vandor, Marie van Vollenhoven, Gerhard Weiss et al.

Harmony in visual compositions is a concept that cannot be defined or easily expressed mathematically, even by humans. The goal of the research described in this paper was to find a numerical representation of artistic compositions with different levels of harmony. We ask humans to rate a collection of grayscale images based on the harmony they convey. To represent the images, a set of special features were designed and extracted. By doing so, it became possible to assign objective measures to subjectively judged compositions. Given the ratings and the extracted features, we utilized machine learning algorithms to evaluate the efficiency of such representations in a harmony classification problem. The best performing model (SVM) achieved 80% accuracy in distinguishing between harmonic and disharmonic images, which reinforces the assumption that concept of harmony can be expressed in a mathematical way that can be assessed by humans.

Rodrigo Alejandro Chávez Mulsa, Gerasimos Spanakis

Recent research in Natural Language Processing has revealed that word embeddings can encode social biases present in the training data which can affect minorities in real world applications. This paper explores the gender bias implicit in Dutch embeddings while investigating whether English language based approaches can also be used in Dutch. We implement the Word Embeddings Association Test (WEAT), Clustering and Sentence Embeddings Association Test (SEAT) methods to quantify the gender bias in Dutch word embeddings, then we proceed to reduce the bias with Hard-Debias and Sent-Debias mitigation methods and finally we evaluate the performance of the debiased embeddings in downstream tasks. The results suggest that, among others, gender bias is present in traditional and contextualized Dutch word embeddings. We highlight how techniques used to measure and reduce bias created for English can be used in Dutch embeddings by adequately translating the data and taking into account the unique characteristics of the language. Furthermore, we analyze the effect of the debiasing techniques on downstream tasks which show a negligible impact on traditional embeddings and a 2% decrease in performance in contextualized embeddings. Finally, we release the translated Dutch datasets to the public along with the traditional embeddings with mitigated bias.

Thalea Schlender, Gerasimos Spanakis

With the vast development and employment of artificial intelligence applications, research into the fairness of these algorithms has been increased. Specifically, in the natural language processing domain, it has been shown that social biases persist in word embeddings and are thus in danger of amplifying these biases when used. As an example of social bias, religious biases are shown to persist in word embeddings and the need for its removal is highlighted. This paper investigates the state-of-the-art multiclass debiasing techniques: Hard debiasing, SoftWEAT debiasing and Conceptor debiasing. It evaluates their performance when removing religious bias on a common basis by quantifying bias removal via the Word Embedding Association Test (WEAT), Mean Average Cosine Similarity (MAC) and the Relative Negative Sentiment Bias (RNSB). By investigating the religious bias removal on three widely used word embeddings, namely: Word2Vec, GloVe, and ConceptNet, it is shown that the preferred method is ConceptorDebiasing. Specifically, this technique manages to decrease the measured religious bias on average by 82,42%, 96,78% and 54,76% for the three word embedding sets respectively.

Danni Liu, Jan Niehues, Gerasimos Spanakis

Automatic speech recognition (ASR) systems are primarily evaluated on transcription accuracy. However, in some use cases such as subtitling, verbatim transcription would reduce output readability given limited screen size and reading time. Therefore, this work focuses on ASR with output compression, a task challenging for supervised approaches due to the scarcity of training data. We first investigate a cascaded system, where an unsupervised compression model is used to post-edit the transcribed speech. We then compare several methods of end-to-end speech recognition under output length constraints. The experiments show that with limited data far less than needed for training a model from scratch, we can adapt a Transformer-based ASR model to incorporate both transcription and compression capabilities. Furthermore, the best performance in terms of WER and ROUGE scores is achieved by explicitly modeling the length constraints within the end-to-end ASR system.

Danni Liu, Gerasimos Spanakis, Jan Niehues

Encoder-decoder models provide a generic architecture for sequence-to-sequence tasks such as speech recognition and translation. While offline systems are often evaluated on quality metrics like word error rates (WER) and BLEU, latency is also a crucial factor in many practical use-cases. We propose three latency reduction techniques for chunk-based incremental inference and evaluate their efficiency in terms of accuracy-latency trade-off. On the 300-hour How2 dataset, we reduce latency by 83% to 0.8 second by sacrificing 1% WER (6% rel.) compared to offline transcription. Although our experiments use the Transformer, the hypothesis selection strategies are applicable to other encoder-decoder models. To avoid expensive re-computation, we use a unidirectionally-attending encoder. After an adaptation procedure to partial sequences, the unidirectional model performs on-par with the original model. We further show that our approach is also applicable to low-latency speech translation. On How2 English-Portuguese speech translation, we reduce latency to 0.7 second (-84% rel.) while incurring a loss of 2.4 BLEU points (5% rel.) compared to the offline system.

Maria Mihaela Trusca, Gerasimos Spanakis

The tiled convolutional neural network (tiled CNN) has been applied only to computer vision for learning invariances. We adjust its architecture to NLP to improve the extraction of the most salient features for sentiment analysis. Knowing that the major drawback of the tiled CNN in the NLP field is its inflexible filter structure, we propose a novel architecture called hybrid tiled CNN that applies a filter only on the words that appear in the similar contexts and on their neighbor words (a necessary step for preventing the loss of some n-grams). The experiments on the datasets of IMDB movie reviews and SemEval 2017 demonstrate the efficiency of the hybrid tiled CNN that performs better than both CNN and tiled CNN.

Cedric Oeldorf, Gerasimos Spanakis

Domains such as logo synthesis, in which the data has a high degree of multi-modality, still pose a challenge for generative adversarial networks (GANs). Recent research shows that progressive training (ProGAN) and mapping network extensions (StyleGAN) enable both increased training stability for higher dimensional problems and better feature separation within the embedded latent space. However, these architectures leave limited control over shaping the output of the network, which is an undesirable trait in the case of logo synthesis. This paper explores a conditional extension to the StyleGAN architecture with the aim of firstly, improving on the low resolution results of previous research and, secondly, increasing the controllability of the output through the use of synthetic class-conditions. Furthermore, methods of extracting such class conditions are explored with a focus on the human interpretability, where the challenge lies in the fact that, by nature, visual logo characteristics are hard to define. The introduced conditional style-based generator architecture is trained on the extracted class-conditions in two experiments and studied relative to the performance of an unconditional model. Results show that, whilst the unconditional model more closely matches the training distribution, high quality conditions enabled the embedding of finer details onto the latent space, leading to more diverse output.

Tobias Bauer, Emre Devrim, Misha Glazunov et al.

Inspired by the recent social movement of #MeToo, we are building a chatbot to assist survivors of sexual harassment cases (designed for the city of Maastricht but can easily be extended). The motivation behind this work is twofold: properly assist survivors of such events by directing them to appropriate institutions that can offer them help and increase the incident documentation so as to gather more data about harassment cases which are currently under reported. We break down the problem into three data science/machine learning components: harassment type identification (treated as a classification problem), spatio-temporal information extraction (treated as Named Entity Recognition problem) and dialogue with the users (treated as a slot-filling based chatbot). We are able to achieve a success rate of more than 98% for the identification of a harassment-or-not case and around 80% for the specific type harassment identification. Locations and dates are identified with more than 90% accuracy and time occurrences prove more challenging with almost 80%. Finally, initial validation of the chatbot shows great potential for the further development and deployment of such a beneficial for the whole society tool.

Matteo Maggiolo, Gerasimos Spanakis

Time Series forecasting (univariate and multivariate) is a problem of high complexity due the different patterns that have to be detected in the input, ranging from high to low frequencies ones. In this paper we propose a new model for timeseries prediction that utilizes convolutional layers for feature extraction, a recurrent encoder and a linear autoregressive component. We motivate the model and we test and compare it against a baseline of widely used existing architectures for univariate and multivariate timeseries. The proposed model appears to outperform the baselines in almost every case of the multivariate timeseries datasets, in some cases even with 50% improvement which shows the strengths of such a hybrid architecture in complex timeseries.

Wouter Leeftink, Gerasimos Spanakis

An obstacle to the development of many natural language processing products is the vast amount of training examples necessary to get satisfactory results. The generation of these examples is often a tedious and time-consuming task. This paper this paper proposes a method to transform the sentiment of sentences in order to limit the work necessary to generate more training data. This means that one sentence can be transformed to an opposite sentiment sentence and should reduce by half the work required in the generation of text. The proposed pipeline consists of a sentiment classifier with an attention mechanism to highlight the short phrases that determine the sentiment of a sentence. Then, these phrases are changed to phrases of the opposite sentiment using a baseline model and an autoencoder approach. Experiments are run on both the separate parts of the pipeline as well as on the end-to-end model. The sentiment classifier is tested on its accuracy and is found to perform adequately. The autoencoder is tested on how well it is able to change the sentiment of an encoded phrase and it was found that such a task is possible. We use human evaluation to judge the performance of the full (end-to-end) pipeline and that reveals that a model using word vectors outperforms the encoder model. Numerical evaluation shows that a success rate of 54.7% is achieved on the sentiment change.

Raffaele Piccini, Gerasimos Spanakis

Conversational agents have begun to rise both in the academic (in terms of research) and commercial (in terms of applications) world. This paper investigates the task of building a non-goal driven conversational agent, using neural network generative models and analyzes how the conversation context is handled. It compares a simpler Encoder-Decoder with a Hierarchical Recurrent Encoder-Decoder architecture, which includes an additional module to model the context of the conversation using previous utterances information. We found that the hierarchical model was able to extract relevant context information and include them in the generation of the output. However, it performed worse (35-40%) than the simple Encoder-Decoder model regarding both grammatically correct output and meaningful response. Despite these results, experiments demonstrate how conversations about similar topics appear close to each other in the context space due to the increased frequency of specific topic-related words, thus leaving promising directions for future research and how the context of a conversation can be exploited.