Giuseppe Riccardi

Michele Yin, Gabriel Roccabruna, Abhinav Azad et al.

The dialogue experience with conversational agents can be greatly enhanced with multimodal and immersive interactions in virtual reality. In this work, we present an open-source architecture with the goal of simplifying the development of conversational agents operating in virtual environments. The architecture offers the possibility of plugging in conversational agents of different domains and adding custom or cloud-based Speech-To-Text and Text-To-Speech models to make the interaction voice-based. Using this architecture, we present two conversational prototypes operating in the digital health domain developed in Unity for both non-immersive displays and VR headsets.

Seyed Mahed Mousavi, Christian Moiola, Massimo Rizzoli et al.

Vision-Language Models (VLMs) are trained on data snapshots of documents, including images and texts. Their training data and evaluation benchmarks are typically static, implicitly treating factual knowledge as time-invariant. However, real-world facts are intrinsically time-sensitive and subject to erratic and periodic changes, causing model predictions to become outdated. We present V-DyKnow, a Visual Dynamic Knowledge benchmark for evaluating time-sensitive factual knowledge in VLMs. Using V-DyKnow, we benchmark closed- and open-source VLMs and analyze a) the reliability (correctness and consistency) of model responses across modalities and input perturbations; b) the efficacy of knowledge editing and multi-modal RAG methods for knowledge updates across modalities; and c) the sources of outdated predictions, through data and mechanistic analysis. Our results show that VLMs frequently output outdated facts, reflecting outdated snapshots used in the (pre-)training phase. Factual reliability degrades from textual to visual stimuli, even when entities are correctly recognized. Besides, existing alignment approaches fail to consistently update the models' knowledge across modalities. Together, these findings highlight fundamental limitations in how current VLMs acquire and update time-sensitive knowledge across modalities. We release the benchmark, code, and evaluation data.

Seyed Mahed Mousavi, Shohei Tanaka, Gabriel Roccabruna et al.

Narratives include a rich source of events unfolding over time and context. Automatic understanding of these events provides a summarised comprehension of the narrative for further computation (such as reasoning). In this paper, we study the Information Status (IS) of the events and propose a novel challenging task: the automatic identification of new events in a narrative. We define an event as a triplet of subject, predicate, and object. The event is categorized as new with respect to the discourse context and whether it can be inferred through commonsense reasoning. We annotated a publicly available corpus of narratives with the new events at sentence level using human annotators. We present the annotation protocol and study the quality of the annotation and the difficulty of the task. We publish the annotated dataset, annotation materials, and machine learning baseline models for the task of new event extraction for narrative understanding.

Seyed Mahed Mousavi, Simone Alghisi, Giuseppe Riccardi

LLMs acquire knowledge from massive data snapshots collected at different timestamps. Their knowledge is then commonly evaluated using static benchmarks. However, factual knowledge is generally subject to time-sensitive changes, and static benchmarks cannot address those cases. We present an approach to dynamically evaluate the knowledge in LLMs and their time-sensitiveness against Wikidata, a publicly available up-to-date knowledge graph. We evaluate the time-sensitive knowledge in twenty-four private and open-source LLMs, as well as the effectiveness of four editing methods in updating the outdated facts. Our results show that 1) outdatedness is a critical problem across state-of-the-art LLMs; 2) LLMs output inconsistent answers when prompted with slight variations of the question prompt; and 3) the performance of the state-of-the-art knowledge editing algorithms is very limited, as they can not reduce the cases of outdatedness and output inconsistency.

Simone Alghisi, Massimo Rizzoli, Seyed Mahed Mousavi et al.

Pointing increases the accuracy and explainability of Large Vision-Language Models (LVLMs) by modeling grounding and reasoning as explicit sequential steps. The model grounds the objects mentioned in the natural-language query by predicting their coordinates, and then generates an answer conditioned on these points. While pointing has been shown to increase LVLMs' accuracy, it is unclear which mechanism supports these gains and its relevance in cognitive tasks. In addition, the reliability of the intermediate points remains understudied, limiting their use as visual explanations. In this work, we study the role of pointing in a cognitive task: zero-shot counting from a visual scene. We fine-tune state-of-the-art LVLMs following two approaches: Direct Counting, where models only predict the total number of objects, and Point-then-Count, where LVLMs generate the target objects' coordinates followed by their count. The results show that Point-then-Count achieves higher out-of-distribution generalization, suggesting that coordinates help LVLMs learn skills rather than overfitting on narrow tasks. Although predicted points are accurately grounded in the image in over 89\% of cases (as measured by F1), performance varies across image regions, revealing spatial biases. Finally, mechanistic analyses show that gains in counting arise from the spatial information encoded in the coordinates.

Seyed Mahed Mousavi, Simone Alghisi, Giuseppe Riccardi

LLMs' sources of knowledge are data snapshots containing factual information about entities collected at different timestamps and from different media types (e.g. wikis, social media, etc.). Such unstructured knowledge is subject to change due to updates through time from past to present. Equally important are the inconsistencies and inaccuracies occurring in different information sources. Consequently, the model's knowledge about an entity may be perturbed while training over the sequence of snapshots or at inference time, resulting in inconsistent and inaccurate model performance. In this work, we study the appropriateness of Large Language Models (LLMs) as repositories of factual knowledge. We consider twenty-four state-of-the-art LLMs that are either closed-, partially (weights), or fully (weight and training data) open-source. We evaluate their reliability in responding to time-sensitive factual questions in terms of accuracy and consistency when prompts are perturbed. We further evaluate the effectiveness of state-of-the-art methods to improve LLMs' accuracy and consistency. We then propose "ENtity-Aware Fine-tuning" (ENAF), a soft neurosymbolic approach aimed at providing a structured representation of entities during fine-tuning to improve the model's performance.

Seyed Mahed Mousavi, Simone Alghisi, Giuseppe Riccardi

Continual Pre-Training (CPT) is widely used for acquiring and updating factual knowledge in LLMs. This practice treats loss as a proxy for knowledge learning, while offering no grounding into how it changes during training. We study CPT as a knowledge learning process rather than a solely optimization problem. We construct a controlled, distribution-matched benchmark of factual documents and interleave diagnostic probes directly into the CPT loop, enabling epoch-level measurement of knowledge acquisition dynamics and changes in Out-Of-Domain (OOD) general skills (e.g., math). We further analyze how CPT reshapes knowledge circuits during training. Across three instruction-tuned LLMs and multiple CPT strategies, optimization and learning systematically diverge as loss decreases monotonically while factual learning is unstable and non-monotonic. Acquired facts are rarely consolidated, learning is strongly conditioned on prior exposure, and OOD performance degrades from early epochs. Circuit analysis reveals rapid reconfiguration of knowledge pathways across epochs, providing an explanation for narrow acquisition windows and systematic forgetting. These results show that loss optimization is misaligned with learning progress in CPT and motivate evaluation of stopping criteria based on task-level learning dynamics.

Massimo Rizzoli, Simone Alghisi, Seyed Mahed Mousavi et al.

Fine-tuning Vision-Language Models (VLMs) is a common strategy to improve performance following an ad-hoc data collection and annotation of real-world scenes. However, this process is often prone to biases, errors, and distribution imbalance, resulting in overfitting and imbalanced performance. Although a few studies have tried to address this problem by generating synthetic data, they lacked control over distribution bias and annotation quality. To address these challenges, we redesign the fine-tuning process in two ways. First, we control the generation of data and its annotations, ensuring it is free from bias, distribution imbalance, and annotation errors. We automatically construct the dataset by comprehensively sampling objects' attributes, including color, shape, size, and position within the scene. Secondly, using this annotated dataset, we fine-tune state-of-the-art VLMs and assess performance transferability to real-world data on the absolute position task. We conduct exhaustive evaluations on both synthetic and real-world benchmarks. Our experiments reveal two key findings: 1) fine-tuning on balanced synthetic data yields uniform performance across the visual scene and mitigates common biases; and 2) fine-tuning on synthetic stimuli significantly improves performance on real-world data (COCO), outperforming models fine-tuned in the matched setting.

Gabriel Roccabruna, Olha Khomyn, Giuseppe Riccardi

AI agents need to plan to achieve complex goals that involve orchestrating perception, sub-goal decomposition, and execution. These plans consist of ordered steps structured according to a Temporal Execution Order (TEO, a directed acyclic graph that ensures each step executes only after its preconditions are satisfied. Existing research on foundational models' understanding of temporal execution is limited to automatically derived annotations, approximations of the TEO as a linear chain, or text-only inputs. To address this gap, we introduce MATEO (MultimodAl Temporal Execution Order), a benchmark designed to assess and improve the temporal reasoning abilities of Large Vision Language Models (LVLMs) required for real-world planning. We acquire a high-quality professional multimodal recipe corpus, authored through a standardized editorial process that decomposes instructions into discrete steps, each paired with corresponding images. We collect TEO annotations as graphs by designing and using a scalable crowdsourcing pipeline. Using MATEO, we evaluate six state-of-the-art LVLMs across model scales, varying language context, multimodal input structure, and fine-tuning strategies.

Seyed Mahed Mousavi, Gabriel Roccabruna, Simone Alghisi et al.

Large Pre-Trained Language Models have demonstrated state-of-the-art performance in different downstream tasks, including dialogue state tracking and end-to-end response generation. Nevertheless, most of the publicly available datasets and benchmarks on task-oriented dialogues focus on written conversations. Consequently, the robustness of the developed models to spoken interactions is unknown. In this work, we have evaluated the performance of LLMs for spoken task-oriented dialogues on the DSTC11 test sets. Due to the lack of proper spoken dialogue datasets, we have automatically transcribed a development set of spoken dialogues with a state-of-the-art ASR engine. We have characterized the ASR-error types and their distributions and simulated these errors in a large dataset of dialogues. We report the intrinsic (perplexity) and extrinsic (human evaluation) performance of fine-tuned GPT-2 and T5 models in two subtasks of response generation and dialogue state tracking, respectively. The results show that LLMs are not robust to spoken noise by default, however, fine-tuning/training such models on a proper dataset of spoken TODs can result in a more robust performance.

Gabriel Roccabruna, Massimo Rizzoli, Giuseppe Riccardi

The automatic detection of temporal relations among events has been mainly investigated with encoder-only models such as RoBERTa. Large Language Models (LLM) have recently shown promising performance in temporal reasoning tasks such as temporal question answering. Nevertheless, recent studies have tested the LLMs' performance in detecting temporal relations of closed-source models only, limiting the interpretability of those results. In this work, we investigate LLMs' performance and decision process in the Temporal Relation Classification task. First, we assess the performance of seven open and closed-sourced LLMs experimenting with in-context learning and lightweight fine-tuning approaches. Results show that LLMs with in-context learning significantly underperform smaller encoder-only models based on RoBERTa. Then, we delve into the possible reasons for this gap by applying explainable methods. The outcome suggests a limitation of LLMs in this task due to their autoregressive nature, which causes them to focus only on the last part of the sequence. Additionally, we evaluate the word embeddings of these two models to better understand their pre-training differences. The code and the fine-tuned models can be found respectively on GitHub.

Seyed Mahed Mousavi, Edoardo Cecchinato, Lucia Hornikova et al.

We conduct a systematic audit of three widely used reasoning benchmarks, SocialIQa, FauxPas-EAI, and ToMi, and uncover pervasive flaws in both benchmark items and evaluation methodology. Using five LLMs (GPT-{3, 3.5, 4, o1}, and LLaMA 3.1) as diagnostic tools, we identify structural, semantic, and pragmatic issues in benchmark design (e.g., duplicated items, ambiguous wording, and implausible answers), as well as scoring procedures that prioritize output form over reasoning process. Through systematic human annotation and re-evaluation on cleaned benchmark subsets, we find that model scores often improve not due to due to erratic surface wording variations and not to improved reasoning. Infact, further analyses show that model performance is highly sensitive to minor input variations such as context availability and phrasing, revealing that high scores may reflect alignment with format-specific cues rather than consistent inference based on the input. These findings challenge the validity of current benchmark-based claims about reasoning in LLMs, and highlight the need for evaluation protocols that assess reasoning as a process of drawing inference from available information, rather than as static output selection. We release audited data and evaluation tools to support more interpretable and diagnostic assessments of model reasoning.

Massimo Rizzoli, Simone Alghisi, Olha Khomyn et al.

While Vision-Language Models (VLMs) have achieved competitive performance in various tasks, their comprehension of the underlying structure and semantics of a scene remains understudied. To investigate the understanding of VLMs, we study their capability regarding object properties and relations in a controlled and interpretable manner. To this scope, we introduce CIVET, a novel and extensible framework for systematiC evaluatIon Via controllEd sTimuli. CIVET addresses the lack of standardized systematic evaluation for assessing VLMs' understanding, enabling researchers to test hypotheses with statistical rigor. With CIVET, we evaluate five state-of-the-art VLMs on exhaustive sets of stimuli, free from annotation noise, dataset-specific biases, and uncontrolled scene complexity. Our findings reveal that 1) current VLMs can accurately recognize only a limited set of basic object properties; 2) their performance heavily depends on the position of the object in the scene; 3) they struggle to understand basic relations among objects. Furthermore, a comparative evaluation with human annotators reveals that VLMs still fall short of achieving human-level accuracy.

Simone Alghisi, Gabriel Roccabruna, Massimo Rizzoli et al.

Vision-Language Models (VLMs) have recently gained attention due to their competitive performance on multiple downstream tasks, achieved by following user-input instructions. However, VLMs still exhibit several limitations in visual reasoning, such as difficulties in identifying relations (e.g., spatial, temporal, and among objects), understanding temporal sequences (e.g., frames), and counting objects. In this work, we go beyond score-level benchmark evaluations of VLMs by investigating the underlying causes of their failures and proposing a targeted approach to improve their reasoning capabilities. We study the reasoning skills of seven state-of-the-art VLMs in the counting task under controlled experimental conditions. Our experiments show that VLMs are highly sensitive to the number and type of objects, their spatial arrangement, and the co-occurrence of distractors. A layer-wise analysis reveals that errors are due to incorrect mapping of the last-layer representation into the output space. Our targeted training shows that fine-tuning just the output layer improves accuracy by up to 21%. We corroborate these findings by achieving consistent improvements on real-world datasets.

Simone Alghisi, Massimo Rizzoli, Gabriel Roccabruna et al.

We study the limitations of Large Language Models (LLMs) for the task of response generation in human-machine dialogue. Several techniques have been proposed in the literature for different dialogue types (e.g., Open-Domain). However, the evaluations of these techniques have been limited in terms of base LLMs, dialogue types and evaluation metrics. In this work, we extensively analyze different LLM adaptation techniques when applied to different dialogue types. We have selected two base LLMs, Llama-2 and Mistral, and four dialogue types Open-Domain, Knowledge-Grounded, Task-Oriented, and Question Answering. We evaluate the performance of in-context learning and fine-tuning techniques across datasets selected for each dialogue type. We assess the impact of incorporating external knowledge to ground the generation in both scenarios of Retrieval-Augmented Generation (RAG) and gold knowledge. We adopt consistent evaluation and explainability criteria for automatic metrics and human evaluation protocols. Our analysis shows that there is no universal best-technique for adapting large language models as the efficacy of each technique depends on both the base LLM and the specific type of dialogue. Last but not least, the assessment of the best adaptation technique should include human evaluation to avoid false expectations and outcomes derived from automatic metrics.

Gabriel Roccabruna, Seyed Mahed Mousavi, Giuseppe Riccardi

The valence analysis of speakers' utterances or written posts helps to understand the activation and variations of the emotional state throughout the conversation. More recently, the concept of Emotion Carriers (EC) has been introduced to explain the emotion felt by the speaker and its manifestations. In this work, we investigate the natural inter-dependency of valence and ECs via a multi-task learning approach. We experiment with Pre-trained Language Models (PLM) for single-task, two-step, and joint settings for the valence and EC prediction tasks. We compare and evaluate the performance of generative (GPT-2) and discriminative (BERT) architectures in each setting. We observed that providing the ground truth label of one task improves the prediction performance of the models in the other task. We further observed that the discriminative model achieves the best trade-off of valence and EC prediction tasks in the joint prediction setting. As a result, we attain a single model that performs both tasks, thus, saving computation resources at training and inference times.

Seyed Mahed Mousavi, Simone Caldarella, Giuseppe Riccardi

Longitudinal Dialogues (LD) are the most challenging type of conversation for human-machine dialogue systems. LDs include the recollections of events, personal thoughts, and emotions specific to each individual in a sparse sequence of dialogue sessions. Dialogue systems designed for LDs should uniquely interact with the users over multiple sessions and long periods of time (e.g. weeks), and engage them in personal dialogues to elaborate on their feelings, thoughts, and real-life events. In this paper, we study the task of response generation in LDs. We evaluate whether general-purpose Pre-trained Language Models (PLM) are appropriate for this purpose. We fine-tune two PLMs, GePpeTto (GPT-2) and iT5, using a dataset of LDs. We experiment with different representations of the personal knowledge extracted from LDs for grounded response generation, including the graph representation of the mentioned events and participants. We evaluate the performance of the models via automatic metrics and the contribution of the knowledge via the Integrated Gradients technique. We categorize the natural language generation errors via human evaluations of contextualization, appropriateness and engagement of the user.

Sebastian P. Bayerl, Gabriel Roccabruna, Shammur Absar Chowdhury et al.

We are interested in the problem of conversational analysis and its application to the health domain. Cognitive Behavioral Therapy is a structured approach in psychotherapy, allowing the therapist to help the patient to identify and modify the malicious thoughts, behavior, or actions. This cooperative effort can be evaluated using the Working Alliance Inventory Observer-rated Shortened - a 12 items inventory covering task, goal, and relationship - which has a relevant influence on therapeutic outcomes. In this work, we investigate the relation between this alliance inventory and the spoken conversations (sessions) between the patient and the psychotherapist. We have delivered eight weeks of e-therapy, collected their audio and video call sessions, and manually transcribed them. The spoken conversations have been annotated and evaluated with WAI ratings by professional therapists. We have investigated speech and language features and their association with WAI items. The feature types include turn dynamics, lexical entrainment, and conversational descriptors extracted from the speech and language signals. Our findings provide strong evidence that a subset of these features are strong indicators of working alliance. To the best of our knowledge, this is the first and a novel study to exploit speech and language for characterising working alliance.

Sebastian P. Bayerl, Aniruddha Tammewar, Korbinian Riedhammer et al.

Personal narratives (PN) - spoken or written - are recollections of facts, people, events, and thoughts from one's own experience. Emotion recognition and sentiment analysis tasks are usually defined at the utterance or document level. However, in this work, we focus on Emotion Carriers (EC) defined as the segments (speech or text) that best explain the emotional state of the narrator ("loss of father", "made me choose"). Once extracted, such EC can provide a richer representation of the user state to improve natural language understanding and dialogue modeling. In previous work, it has been shown that EC can be identified using lexical features. However, spoken narratives should provide a richer description of the context and the users' emotional state. In this paper, we leverage word-based acoustic and textual embeddings as well as early and late fusion techniques for the detection of ECs in spoken narratives. For the acoustic word-level representations, we use Residual Neural Networks (ResNet) pretrained on separate speech emotion corpora and fine-tuned to detect EC. Experiments with different fusion and system combination strategies show that late fusion leads to significant improvements for this task.

Juan Manuel Mayor-Torres, Sara Medina-DeVilliers, Tessa Clarkson et al.

Current models on Explainable Artificial Intelligence (XAI) have shown an evident and quantified lack of reliability for measuring feature-relevance when statistically entangled features are proposed for training deep classifiers. There has been an increase in the application of Deep Learning in clinical trials to predict early diagnosis of neuro-developmental disorders, such as Autism Spectrum Disorder (ASD). However, the inclusion of more reliable saliency-maps to obtain more trustworthy and interpretable metrics using neural activity features is still insufficiently mature for practical applications in diagnostics or clinical trials. Moreover, in ASD research the inclusion of deep classifiers that use neural measures to predict viewed facial emotions is relatively unexplored. Therefore, in this study we propose the evaluation of a Convolutional Neural Network (CNN) for electroencephalography (EEG)-based facial emotion recognition decoding complemented with a novel RemOve-And-Retrain (ROAR) methodology to recover highly relevant features used in the classifier. Specifically, we compare well-known relevance maps such as Layer-Wise Relevance Propagation (LRP), PatternNet, Pattern-Attribution, and Smooth-Grad Squared. This study is the first to consolidate a more transparent feature-relevance calculation for a successful EEG-based facial emotion recognition using a within-subject-trained CNN in typically-developed and ASD individuals.

Juan Manuel Mayor-Torres, Mirco Ravanelli, Sara E. Medina-DeVilliers et al.

Machine learning methods, such as deep learning, show promising results in the medical domain. However, the lack of interpretability of these algorithms may hinder their applicability to medical decision support systems. This paper studies an interpretable deep learning technique, called SincNet. SincNet is a convolutional neural network that efficiently learns customized band-pass filters through trainable sinc-functions. In this study, we use SincNet to analyze the neural activity of individuals with Autism Spectrum Disorder (ASD), who experience characteristic differences in neural oscillatory activity. In particular, we propose a novel SincNet-based neural network for detecting emotions in ASD patients using EEG signals. The learned filters can be easily inspected to detect which part of the EEG spectrum is used for predicting emotions. We found that our system automatically learns the high-$α$ (9-13 Hz) and $β$ (13-30 Hz) band suppression often present in individuals with ASD. This result is consistent with recent neuroscience studies on emotion recognition, which found an association between these band suppressions and the behavioral deficits observed in individuals with ASD. The improved interpretability of SincNet is achieved without sacrificing performance in emotion recognition.

Aniruddha Tammewar, Alessandra Cervone, Giuseppe Riccardi

Personal Narratives (PN) - recollections of facts, events, and thoughts from one's own experience - are often used in everyday conversations. So far, PNs have mainly been explored for tasks such as valence prediction or emotion classification (e.g. happy, sad). However, these tasks might overlook more fine-grained information that could prove to be relevant for understanding PNs. In this work, we propose a novel task for Narrative Understanding: Emotion Carrier Recognition (ECR). Emotion carriers, the text fragments that carry the emotions of the narrator (e.g. loss of a grandpa, high school reunion), provide a fine-grained description of the emotion state. We explore the task of ECR in a corpus of PNs manually annotated with emotion carriers and investigate different machine learning models for the task. We propose evaluation strategies for ECR including metrics that can be appropriate for different tasks.

Alessandra Cervone, Giuseppe Riccardi

In this work, we investigate the human perception of coherence in open-domain dialogues. In particular, we address the problem of annotating and modeling the coherence of next-turn candidates while considering the entire history of the dialogue. First, we create the Switchboard Coherence (SWBD-Coh) corpus, a dataset of human-human spoken dialogues annotated with turn coherence ratings, where next-turn candidate utterances ratings are provided considering the full dialogue context. Our statistical analysis of the corpus indicates how turn coherence perception is affected by patterns of distribution of entities previously introduced and the Dialogue Acts used. Second, we experiment with different architectures to model entities, Dialogue Acts and their combination and evaluate their performance in predicting human coherence ratings on SWBD-Coh. We find that models combining both DA and entity information yield the best performances both for response selection and turn coherence rating.

Aniruddha Tammewar, Alessandra Cervone, Eva-Maria Messner et al.

We are interested in the problem of understanding personal narratives (PN) - spoken or written - recollections of facts, events, and thoughts. In PN, emotion carriers are the speech or text segments that best explain the emotional state of the user. Such segments may include entities, verb or noun phrases. Advanced automatic understanding of PNs requires not only the prediction of the user emotional state but also to identify which events (e.g. "the loss of relative" or "the visit of grandpa") or people ( e.g. "the old group of high school mates") carry the emotion manifested during the personal recollection. This work proposes and evaluates an annotation model for identifying emotion carriers in spoken personal narratives. Compared to other text genres such as news and microblogs, spoken PNs are particularly challenging because a narrative is usually unstructured, involving multiple sub-events and characters as well as thoughts and associated emotions perceived by the narrator. In this work, we experiment with annotating emotion carriers from speech transcriptions in the Ulm State-of-Mind in Speech (USoMS) corpus, a dataset of German PNs. We believe this resource could be used for experiments in the automatic extraction of emotion carriers from PN, a task that could provide further advancements in narrative understanding.

Giuliano Tortoreto, Evgeny A. Stepanov, Alessandra Cervone et al.

The increase in the prevalence of mental health problems has coincided with a growing popularity of health related social networking sites. Regardless of their therapeutic potential, On-line Support Groups (OSGs) can also have negative effects on patients. In this work we propose a novel methodology to automatically verify the presence of therapeutic factors in social networking websites by using Natural Language Processing (NLP) techniques. The methodology is evaluated on On-line asynchronous multi-party conversations collected from an OSG and Twitter. The results of the analysis indicate that therapeutic factors occur more frequently in OSG conversations than in Twitter conversations. Moreover, the analysis of OSG conversations reveals that the users of that platform are supportive, and interactions are likely to lead to the improvement of their emotional state. We believe that our method provides a stepping stone towards automatic analysis of emotional states of users of online platforms. Possible applications of the method include provision of guidelines that highlight potential implications of using such platforms on users' mental health, and/or support in the analysis of their impact on specific individuals.

Federico Marinelli, Alessandra Cervone, Giuliano Tortoreto et al.

Natural Language Understanding (NLU) models are typically trained in a supervised learning framework. In the case of intent classification, the predicted labels are predefined and based on the designed annotation schema while the labelling process is based on a laborious task where annotators manually inspect each utterance and assign the corresponding label. We propose an Active Annotation (AA) approach where we combine an unsupervised learning method in the embedding space, a human-in-the-loop verification process, and linguistic insights to create lexicons that can be open categories and adapted over time. In particular, annotators define the y-label space on-the-fly during the annotation using an iterative process and without the need for prior knowledge about the input data. We evaluate the proposed annotation paradigm in a real use-case NLU scenario. Results show that our Active Annotation paradigm achieves accurate and higher quality training data, with an annotation speed of an order of magnitude higher with respect to the traditional human-only driven baseline annotation methodology.

Andrei C. Coman, Koichiro Yoshino, Yukitoshi Murase et al.

In a human-machine dialog scenario, deciding the appropriate time for the machine to take the turn is an open research problem. In contrast, humans engaged in conversations are able to timely decide when to interrupt the speaker for competitive or non-competitive reasons. In state-of-the-art turn-by-turn dialog systems the decision on the next dialog action is taken at the end of the utterance. In this paper, we propose a token-by-token prediction of the dialog state from incremental transcriptions of the user utterance. To identify the point of maximal understanding in an ongoing utterance, we a) implement an incremental Dialog State Tracker which is updated on a token basis (iDST) b) re-label the Dialog State Tracking Challenge 2 (DSTC2) dataset and c) adapt it to the incremental turn-taking experimental scenario. The re-labeling consists of assigning a binary value to each token in the user utterance that allows to identify the appropriate point for taking the turn. Finally, we implement an incremental Turn Taking Decider (iTTD) that is trained on these new labels for the turn-taking decision. We show that the proposed model can achieve a better performance compared to a deterministic handcrafted turn-taking algorithm.

Aniruddha Tammewar, Alessandra Cervone, Eva-Maria Messner et al.

Automated prediction of valence, one key feature of a person's emotional state, from individuals' personal narratives may provide crucial information for mental healthcare (e.g. early diagnosis of mental diseases, supervision of disease course, etc.). In the Interspeech 2018 ComParE Self-Assessed Affect challenge, the task of valence prediction was framed as a three-class classification problem using 8 seconds fragments from individuals' narratives. As such, the task did not allow for exploring contextual information of the narratives. In this work, we investigate the intrinsic information from multiple narratives recounted by the same individual in order to predict their current state-of-mind. Furthermore, with generalizability in mind, we decided to focus our experiments exclusively on textual information as the public availability of audio narratives is limited compared to text. Our hypothesis is, that context modeling might provide insights about emotion triggering concepts (e.g. events, people, places) mentioned in the narratives that are linked to an individual's state of mind. We explore multiple machine learning techniques to model narratives. We find that the models are able to capture inter-individual differences, leading to more accurate predictions of an individual's emotional state, as compared to single narratives.

Jacopo Gobbi, Evgeny Stepanov, Giuseppe Riccardi

Concept tagging is a type of structured learning needed for natural language understanding (NLU) systems. In this task, meaning labels from a domain ontology are assigned to word sequences. In this paper, we review the algorithms developed over the last twenty five years. We perform a comparative evaluation of generative, discriminative and deep learning methods on two public datasets. We report on the statistical variability performance measurements. The third contribution is the release of a repository of the algorithms, datasets and recipes for NLU evaluation.

Alessandra Cervone, Evgeny Stepanov, Giuseppe Riccardi

Coherence across multiple turns is a major challenge for state-of-the-art dialogue models. Arguably the most successful approach to automatically learning text coherence is the entity grid, which relies on modelling patterns of distribution of entities across multiple sentences of a text. Originally applied to the evaluation of automatic summaries and the news genre, among its many extensions, this model has also been successfully used to assess dialogue coherence. Nevertheless, both the original grid and its extensions do not model intents, a crucial aspect that has been studied widely in the literature in connection to dialogue structure. We propose to augment the original grid document representation for dialogue with the intentional structure of the conversation. Our models outperform the original grid representation on both text discrimination and insertion, the two main standard tasks for coherence assessment across three different dialogue datasets, confirming that intents play a key role in modelling dialogue coherence.

Stefano Mezza, Alessandra Cervone, Giuliano Tortoreto et al.

Dialogue Act (DA) tagging is crucial for spoken language understanding systems, as it provides a general representation of speakers' intents, not bound to a particular dialogue system. Unfortunately, publicly available data sets with DA annotation are all based on different annotation schemes and thus incompatible with each other. Moreover, their schemes often do not cover all aspects necessary for open-domain human-machine interaction. In this paper, we propose a methodology to map several publicly available corpora to a subset of the ISO standard, in order to create a large task-independent training corpus for DA classification. We show the feasibility of using this corpus to train a domain-independent DA tagger testing it on out-of-domain conversational data, and argue the importance of training on multiple corpora to achieve robustness across different DA categories.

Evgeny Stepanov, Stephane Lathuiliere, Shammur Absar Chowdhury et al.

Depression is a major debilitating disorder which can affect people from all ages. With a continuous increase in the number of annual cases of depression, there is a need to develop automatic techniques for the detection of the presence and extent of depression. In this AVEC challenge we explore different modalities (speech, language and visual features extracted from face) to design and develop automatic methods for the detection of depression. In psychology literature, the PHQ-8 questionnaire is well established as a tool for measuring the severity of depression. In this paper we aim to automatically predict the PHQ-8 scores from features extracted from the different modalities. We show that visual features extracted from facial landmarks obtain the best performance in terms of estimating the PHQ-8 results with a mean absolute error (MAE) of 4.66 on the development set. Behavioral characteristics from speech provide an MAE of 4.73. Language features yield a slightly higher MAE of 5.17. When switching to the test set, our Turn Features derived from audio transcriptions achieve the best performance, scoring an MAE of 4.11 (corresponding to an RMSE of 4.94), which makes our system the winner of the AVEC 2017 depression sub-challenge.

Firoj Alam, Morena Danieli, Giuseppe Riccardi

Empathy, as defined in behavioral sciences, expresses the ability of human beings to recognize, understand and react to emotions, attitudes and beliefs of others. The lack of an operational definition of empathy makes it difficult to measure it. In this paper, we address two related problems in automatic affective behavior analysis: the design of the annotation protocol and the automatic recognition of empathy from spoken conversations. We propose and evaluate an annotation scheme for empathy inspired by the modal model of emotions. The annotation scheme was evaluated on a corpus of real-life, dyadic spoken conversations. In the context of behavioral analysis, we designed an automatic segmentation and classification system for empathy. Given the different speech and language levels of representation where empathy may be communicated, we investigated features derived from the lexical and acoustic spaces. The feature development process was designed to support both the fusion and automatic selection of relevant features from high dimensional space. The automatic classification system was evaluated on call center conversations where it showed significantly better performance than the baseline.