Gabriel Roccabruna

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, 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, 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.

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.