A virtual agent to sustaining children's engagement in language learning

In this thesis we aimed to explore the potential of gamification - defined as “the use of game elements in non-game contexts” [30] - in increasing children's (aged 5 to 6) engagement with the task. This is mainly due to the fact that our world is living a technological era, and videogames are an example of this engagement by being able to maintain children’s (and adults) engagement for hours straight. For the purpose of limiting complexity, we only addressed the feedback element by introducing it with an anthropomorphic virtual agent (human-like aspect), because research shows that virtual agents (VA’s) can influence behavioural change [17], or even induce emotions on humans both through the use of feedback provided and their facial expressions, which can interpreted in the same way as of humans’ [2]. By pairing the VA with the gamification concept, we wanted to 1) create a VA that is likely to be well-received by children (appearance and behaviour), and 2) have the immediate feedback that games have, so we can give children an assessment of their actions in real-time, as opposed to waiting for feedback from someone (traditional teaching), and with this give students more chances to succeed [32, 43]. Our final system consisted on a virtual environment, where children formed words that corresponded to a given image. In order to measure the impact that the VA had on engagement, the system was developed in two versions: one version of the system was limited to provide a simple feedback environment, where the VA provided feedback, by responding with simple phrases (i.e. “correct” or “incorrect”); for the second version, the VA had a more complex approach where it tried to encourage children to complete the word – a motivational feedback - even when they weren’t succeeding. Lastly we conducted a field study with two groups of children, where one group tested the version with the simple feedback, and the other group tested the ‘motivational’ version of the system. We used a quantitative approach to analyze the collected data that measured the engagement, based on the number of tasks (words) completed and time spent with system. The results of the evaluation showed that the use of motivational feedback may carry a positive effect on engaging children.

Advanced neuroimaging methodologies to improve connectivity detection in normal and abnormal language brain networks

The language connectome was in-vivo investigated using multimodal non-invasive quantitative MRI. In PPA patients (n=18) recruited by the IRCCS ISNB, Bologna, cortical thickness measures showed a predominant reduction on the left hemisphere (p<0.005) with respect to matched healthy controls (HC) (n=18), and an accuracy of 86.1% in discrimination from Alzheimer’s disease patients (n=18). The left temporal and para-hippocampal gyri significantly correlated (p<0.01) with language fluency. In PPA patients (n=31) recruited by the Northwestern University Chicago, DTI measures were longitudinally evaluated (2-years follow-up) under the supervision of Prof. M. Catani, King’s College London. Significant differences with matched HC (n=27) were found, tract-localized at baseline and widespread in the follow-up. Language assessment scores correlated with arcuate (AF) and uncinate (UF) fasciculi DTI measures. In left-ischemic stroke patients (n=16) recruited by the NatBrainLab, King’s College London, language recovery was longitudinally evaluated (6-months follow-up). Using arterial spin labelling imaging a significant correlation (p<0.01) between language recovery and cerebral blood flow asymmetry, was found in the middle cerebral artery perfusion, towards the right. In HC (n=29) recruited by the DIBINEM Functional MR Unit, University of Bologna, an along-tract algorithm was developed suitable for different tractography methods, using the Laplacian operator. A higher left superior temporal gyrus and precentral operculum AF connectivity was found (Talozzi L et al., 2018), and lateralized UF projections towards the left dorsal orbital cortex. In HC (n=50) recruited in the Human Connectome Project, a new tractography-driven approach was developed for left association fibres, using a principal component analysis. The first component discriminated cortical areas typically connected by the AF, suggesting a good discrimination of cortical areas sharing a similar connectivity pattern. The evaluation of morphological, microstructural and metabolic measures could be used as in-vivo biomarkers to monitor language impairment related to neurodegeneration or as surrogate of cognitive rehabilitation/interventional treatment efficacy.

Affordances and Language: How the Level of Object Familiarity modulates the Manipulation and Categorization of Objects across Development

According to much evidence, observing objects activates two types of information: structural properties, i.e., the visual information about the structural features of objects, and function knowledge, i.e., the conceptual information about their skilful use. Many studies so far have focused on the role played by these two kinds of information during object recognition and on their neural underpinnings. However, to the best of our knowledge no study so far has focused on the different activation of this information (structural vs. function) during object manipulation and conceptualization, depending on the age of participants and on the level of object familiarity (familiar vs. non-familiar). Therefore, the main aim of this dissertation was to investigate how actions and concepts related to familiar and non-familiar objects may vary across development. To pursue this aim, four studies were carried out. A first study led to the creation of the Familiar and Non-Familiar Stimuli Database, a set of everyday objects classified by Italian pre-schoolers, schoolers, and adults, useful to verify how object knowledge is modulated by age and frequency of use. A parallel study demonstrated that factors such as sociocultural dynamics may affect the perception of objects. Specifically, data for familiarity, naming, function, using and frequency of use of the objects used to create the Familiar And Non-Familiar Stimuli Database were collected with Dutch and Croatian children and adults. The last two studies on object interaction and language provide further evidence in support of the literature on affordances and on the link between affordances and the cognitive process of language from a developmental point of view, supporting the perspective of a situated cognition and emphasizing the crucial role of human experience.

Apprendimento non supervisionato di rappresentazioni e legami di similarità tra eventi menzionati nella letteratura biomedica

L'estrazione automatica degli eventi biomedici dalla letteratura scientifica ha catturato un forte interesse nel corso degli ultimi anni, dimostrandosi in grado di riconoscere interazioni complesse e semanticamente ricche espresse all'interno del testo. Purtroppo però, esistono davvero pochi lavori focalizzati sull'apprendimento di embedding o di metriche di similarità per i grafi evento. Questa lacuna lascia le relazioni biologiche scollegate, impedendo l'applicazione di tecniche di machine learning che potrebbero dare un importante contributo al progresso scientifico. Approfittando dei vantaggi delle recenti soluzioni di deep graph kernel e dei language model preaddestrati, proponiamo Deep Divergence Event Graph Kernels (DDEGK), un metodo non supervisionato e induttivo in grado di mappare gli eventi all'interno di uno spazio vettoriale, preservando le loro similarità semantiche e strutturali. Diversamente da molti altri sistemi, DDEGK lavora a livello di grafo e non richiede nè etichette e feature specifiche per un determinato task, nè corrispondenze note tra i nodi. A questo scopo, la nostra soluzione mette a confronto gli eventi con un piccolo gruppo di eventi prototipo, addestra delle reti di cross-graph attention per andare a individuare i legami di similarità tra le coppie di nodi (rafforzando l'interpretabilità), e impiega dei modelli basati su transformer per la codifica degli attributi continui. Sono stati fatti ampi esperimenti su dieci dataset biomedici. Mostriamo che le nostre rappresentazioni possono essere utilizzate in modo efficace in task quali la classificazione di grafi, clustering e visualizzazione e che, allo stesso tempo, sono in grado di semplificare il task di semantic textual similarity. Risultati empirici dimostrano che DDEGK supera significativamente gli altri modelli che attualmente detengono lo stato dell'arte.

Self-supervised information retrieval: a novel approach based on Deep Metric Learning and Neural Language Models

Most of the existing open-source search engines, utilize keyword or tf-idf based techniques to find relevant documents and web pages relative to an input query. Although these methods, with the help of a page rank or knowledge graphs, proved to be effective in some cases, they often fail to retrieve relevant instances for more complicated queries that would require a semantic understanding to be exploited. In this Thesis, a self-supervised information retrieval system based on transformers is employed to build a semantic search engine over the library of Gruppo Maggioli company. Semantic search or search with meaning can refer to an understanding of the query, instead of simply finding words matches and, in general, it represents knowledge in a way suitable for retrieval. We chose to investigate a new self-supervised strategy to handle the training of unlabeled data based on the creation of pairs of ’artificial’ queries and the respective positive passages. We claim that by removing the reliance on labeled data, we may use the large volume of unlabeled material on the web without being limited to languages or domains where labeled data is abundant.

Previsione del Successo di Prodotti di Moda Prima della Commercializzazione: un Nuovo Dataset e Modello di Vision-Language Transformer

A differenza di quanto avviene nel commercio tradizionale, in quello online il cliente non ha la possibilità di toccare con mano o provare il prodotto. La decisione di acquisto viene maturata in base ai dati messi a disposizione dal venditore attraverso titolo, descrizioni, immagini e alle recensioni di clienti precedenti. É quindi possibile prevedere quanto un prodotto venderà sulla base di queste informazioni. La maggior parte delle soluzioni attualmente presenti in letteratura effettua previsioni basandosi sulle recensioni, oppure analizzando il linguaggio usato nelle descrizioni per capire come questo influenzi le vendite. Le recensioni, tuttavia, non sono informazioni note ai venditori prima della commercializzazione del prodotto; usando solo dati testuali, inoltre, si tralascia l’influenza delle immagini. L'obiettivo di questa tesi è usare modelli di machine learning per prevedere il successo di vendita di un prodotto a partire dalle informazioni disponibili al venditore prima della commercializzazione. Si fa questo introducendo un modello cross-modale basato su Vision-Language Transformer in grado di effettuare classificazione. Un modello di questo tipo può aiutare i venditori a massimizzare il successo di vendita dei prodotti. A causa della mancanza, in letteratura, di dataset contenenti informazioni relative a prodotti venduti online che includono l’indicazione del successo di vendita, il lavoro svolto comprende la realizzazione di un dataset adatto a testare la soluzione sviluppata. Il dataset contiene un elenco di 78300 prodotti di Moda venduti su Amazon, per ognuno dei quali vengono riportate le principali informazioni messe a disposizione dal venditore e una misura di successo sul mercato. Questa viene ricavata a partire dal gradimento espresso dagli acquirenti e dal posizionamento del prodotto in una graduatoria basata sul numero di esemplari venduti.

End-to-end Deep Metric Learning con Vision-Language Model per il Fashion Image Captioning

L'image captioning è un task di machine learning che consiste nella generazione di una didascalia, o caption, che descriva le caratteristiche di un'immagine data in input. Questo può essere applicato, ad esempio, per descrivere in dettaglio i prodotti in vendita su un sito di e-commerce, migliorando l'accessibilità del sito web e permettendo un acquisto più consapevole ai clienti con difficoltà visive. La generazione di descrizioni accurate per gli articoli di moda online è importante non solo per migliorare le esperienze di acquisto dei clienti, ma anche per aumentare le vendite online. Oltre alla necessità di presentare correttamente gli attributi degli articoli, infatti, descrivere i propri prodotti con il giusto linguaggio può contribuire a catturare l'attenzione dei clienti. In questa tesi, ci poniamo l'obiettivo di sviluppare un sistema in grado di generare una caption che descriva in modo dettagliato l'immagine di un prodotto dell'industria della moda dato in input, sia esso un capo di vestiario o un qualche tipo di accessorio. A questo proposito, negli ultimi anni molti studi hanno proposto soluzioni basate su reti convoluzionali e LSTM. In questo progetto proponiamo invece un'architettura encoder-decoder, che utilizza il modello Vision Transformer per la codifica delle immagini e GPT-2 per la generazione dei testi. Studiamo inoltre come tecniche di deep metric learning applicate in end-to-end durante l'addestramento influenzino le metriche e la qualità delle caption generate dal nostro modello.

Fuori dai binari. Tradurre le identita non binarie nel romanzo Young Adult I Wish You All the Best di Mason Deaver

The goal of this final dissertation is to propose a partial translation of the Young Adult novel I Wish You All the Best by nonbinary author Mason Deaver, which was published in the United States by PUSH in 2019 and is currently still unpublished in Italy. The book follows the story of eighteen-year-old Ben De Backer. After coming out as nonbinary to their parents, they are kicked out of their home; thanks to the help of their older sister, they will manage to get back on their feet and start a new life, although their trauma will continue to pose a great obstacle to their growth and happiness. I Wish You All the Best is one of the first novels with a nonbinary protagonist in the Young Adult category and is particularly interesting from a linguistic standpoint. The first chapter serves as introduction to the author, Mason Deaver, their life and their relationship with gender identity and writing; and also as an introduction to the book itself and how its main characters and themes are developed in relation to the plot. The second chapter looks at the Young Adult genre as a whole; it explores its origins and development through the years and seeks to find a definition for this type of literature which has often been misrepresented both in the eyes of literary critics and the general public. Additionally, the chapter will describe some of the main characteristics of YA literature and analyze them in relation to I Wish You All the Best. The third chapter aims to examine the role of language, and how it can be inclusive to people outside of the gender binary. It will also touch upon the sociocultural contexts in which the speakers of different languages operate. The fourth and final chapter will present the translation of several passages taken from the novel I Wish You All the Best and offer an analysis of the most interesting and difficult aspects of this translation process.

Mocku for change: an explorative study on the use of “mockumentary” as a new language for climate change education

In the ‘society of acceleration and uncertainty’ (Rosa, 2013), the young are struggling to interpret our complex and fast-changing world. The entity and the velocity of changes are so enormous that we need new narratives or languages to conceptualise them. Among those changes “Climate Change” is placed in a particular difficult position. As the writer A. Ghosh said: “The current climate crisis is also a crisis of culture, and thus of the imagination”. In fact, in today’s literature and cinema a strong dichotomy exists between fictional and non-fictional works, but none of those extremities seems suitable to picture an “adequate representation” of climate change issues, particularly those that are related to future. The main goal of my study, carried out within FEDORA EU project, was to understand to what extent the hybrid film form called “mockumentary” (a language that adopts the aesthetics of factual production to give an illusion of truth to invented stories) could inspire and help students in overcoming the mentioned dichotomy, working as a tool to foster the development of argumentative and imaginative skills needed to picture “immaginary yet realistic” climate change scenarios.

Vocabulary Transfer and Knowledge Distillation for Language Model Compression

Natural Language Processing (NLP) has seen tremendous improvements over the last few years. Transformer architectures achieved impressive results in almost any NLP task, such as Text Classification, Machine Translation, and Language Generation. As time went by, transformers continued to improve thanks to larger corpora and bigger networks, reaching hundreds of billions of parameters. Training and deploying such large models has become prohibitively expensive, such that only big high tech companies can afford to train those models. Therefore, a lot of research has been dedicated to reducing a model’s size. In this thesis, we investigate the effects of Vocabulary Transfer and Knowledge Distillation for compressing large Language Models. The goal is to combine these two methodologies to further compress models without significant loss of performance. In particular, we designed different combination strategies and conducted a series of experiments on different vertical domains (medical, legal, news) and downstream tasks (Text Classification and Named Entity Recognition). Four different methods involving Vocabulary Transfer (VIPI) with and without a Masked Language Modelling (MLM) step and with and without Knowledge Distillation are compared against a baseline that assigns random vectors to new elements of the vocabulary. Results indicate that VIPI effectively transfers information of the original vocabulary and that MLM is beneficial. It is also noted that both vocabulary transfer and knowledge distillation are orthogonal to one another and may be applied jointly. The application of knowledge distillation first before subsequently applying vocabulary transfer is recommended. Finally, model performance due to vocabulary transfer does not always show a consistent trend as the vocabulary size is reduced. Hence, the choice of vocabulary size should be empirically selected by evaluation on the downstream task similar to hyperparameter tuning.

Audiovisual materials and second-language acquisition: a comparison between native language subtitles and target language captions.

Over the last few years, the massive popularity of video streaming platforms has managed to impact our daily habits by making the watching of movies and TV shows one of the main activities of our free time. By providing a wide range of foreign language audiovisual content, these entertainment services may represent a powerful resource for language learners, as they provide them with the possibility to be exposed to authentic input. Moreover, research has shown the beneficial role of audiovisual textual aids such as native language subtitles and target language captions in enhancing language skills such as vocabulary and listening comprehension. The aim of this thesis is to analyze the existing literature on the subject of subtitled and captioned audiovisual materials used as a pedagogical tool for informal language learning.

Graph-based Vision-Language Model for Multi-Modal Retrieval of Fashion Products

Artificial Intelligence is reshaping the field of fashion industry in different ways. E-commerce retailers exploit their data through AI to enhance their search engines, make outfit suggestions and forecast the success of a specific fashion product. However, it is a challenging endeavour as the data they possess is huge, complex and multi-modal. The most common way to search for fashion products online is by matching keywords with phrases in the product's description which are often cluttered, inadequate and differ across collections and sellers. A customer may also browse an online store's taxonomy, although this is time-consuming and doesn't guarantee relevant items. With the advent of Deep Learning architectures, particularly Vision-Language models, ad-hoc solutions have been proposed to model both the product image and description to solve this problems. However, the suggested solutions do not exploit effectively the semantic or syntactic information of these modalities, and the unique qualities and relations of clothing items. In this work of thesis, a novel approach is proposed to address this issues, which aims to model and process images and text descriptions as graphs in order to exploit the relations inside and between each modality and employs specific techniques to extract syntactic and semantic information. The results obtained show promising performances on different tasks when compared to the present state-of-the-art deep learning architectures.

Egfr Activating Mutations And Their Association With Response To Platinum-doublet Chemotherapy In Brazilian Non-small Cell Lung Cancer Patients.

The aim of the study was to analyze the frequency of epidermal growth factor receptor (EGFR) mutations in Brazilian non-small cell lung cancer patients and to correlate these mutations with response to benefit of platinum-based chemotherapy in non-small cell lung cancer (NSCLC). Our cohort consisted of prospective patients with NSCLCs who received chemotherapy (platinum derivates plus paclitaxel) at the [UNICAMP], Brazil. EGFR exons 18-21 were analyzed in tumor-derived DNA. Fifty patients were included in the study (25 with adenocarcinoma). EGFR mutations were identified in 6/50 (12 %) NSCLCs and in 6/25 (24 %) adenocarcinomas; representing the frequency of EGFR mutations in a mostly self-reported White (82.0 %) southeastern Brazilian population of NSCLCs. Patients with NSCLCs harboring EGFR exon 19 deletions or the exon 21 L858R mutation were found to have a higher chance of response to platinum-paclitaxel (OR 9.67 [95 % CI 1.03-90.41], p = 0.047). We report the frequency of EGFR activating mutations in a typical southeastern Brazilian population with NSCLC, which are similar to that of other countries with Western European ethnicity. EGFR mutations seem to be predictive of a response to platinum-paclitaxel, and additional studies are needed to confirm or refute this relationship.

Delays In Receiving Obstetric Care And Poor Maternal Outcomes: Results From A National Multicentre Cross-sectional Study.

The vast majority of maternal deaths in low-and middle-income countries are preventable. Delay in obtaining access to appropriate health care is a fairly common problem which can be improved. The objective of this study was to explore the association between delay in providing obstetric health care and severe maternal morbidity/death. This was a multicentre cross-sectional study, involving 27 referral obstetric facilities in all Brazilian regions between 2009 and 2010. All women admitted to the hospital with a pregnancy-related cause were screened, searching for potentially life-threatening conditions (PLTC), maternal death (MD) and maternal near-miss (MNM) cases, according to the WHO criteria. Data on delays were collected by medical chart review and interview with the medical staff. The prevalence of the three different types of delays was estimated according to the level of care and outcome of the complication. For factors associated with any delay, the PR and 95%CI controlled for cluster design were estimated. A total of 82,144 live births were screened, with 9,555 PLTC, MNM or MD cases prospectively identified. Overall, any type of delay was observed in 53.8% of cases; delay related to user factors was observed in 10.2%, 34.6% of delays were related to health service accessibility and 25.7% were related to quality of medical care. The occurrence of any delay was associated with increasing severity of maternal outcome: 52% in PLTC, 68.4% in MNM and 84.1% in MD. Although this was not a population-based study and the results could not be generalized, there was a very clear and significant association between frequency of delay and severity of outcome, suggesting that timely and proper management are related to survival.

Fatty Acid Synthase Inhibitors Induce Apoptosis In Non-tumorigenic Melan-a Cells Associated With Inhibition Of Mitochondrial Respiration.

The metabolic enzyme fatty acid synthase (FASN) is responsible for the endogenous synthesis of palmitate, a saturated long-chain fatty acid. In contrast to most normal tissues, a variety of human cancers overexpress FASN. One such cancer is cutaneous melanoma, in which the level of FASN expression is associated with tumor invasion and poor prognosis. We previously reported that two FASN inhibitors, cerulenin and orlistat, induce apoptosis in B16-F10 mouse melanoma cells via the intrinsic apoptosis pathway. Here, we investigated the effects of these inhibitors on non-tumorigenic melan-a cells. Cerulenin and orlistat treatments were found to induce apoptosis and decrease cell proliferation, in addition to inducing the release of mitochondrial cytochrome c and activating caspases-9 and -3. Transfection with FASN siRNA did not result in apoptosis. Mass spectrometry analysis demonstrated that treatment with the FASN inhibitors did not alter either the mitochondrial free fatty acid content or composition. This result suggests that cerulenin- and orlistat-induced apoptosis events are independent of FASN inhibition. Analysis of the energy-linked functions of melan-a mitochondria demonstrated the inhibition of respiration, followed by a significant decrease in mitochondrial membrane potential (ΔΨm) and the stimulation of superoxide anion generation. The inhibition of NADH-linked substrate oxidation was approximately 40% and 61% for cerulenin and orlistat treatments, respectively, and the inhibition of succinate oxidation was approximately 46% and 52%, respectively. In contrast, no significant inhibition occurred when respiration was supported by the complex IV substrate N,N,N',N'-tetramethyl-p-phenylenediamine (TMPD). The protection conferred by the free radical scavenger N-acetyl-cysteine indicates that the FASN inhibitors induced apoptosis through an oxidative stress-associated mechanism. In combination, the present results demonstrate that cerulenin and orlistat induce apoptosis in non-tumorigenic cells via mitochondrial dysfunction, independent of FASN inhibition.