Bitext alignment: building and evaluating a bilingual corpus and translation memory of academic course descriptions

Following the internationalization of contemporary higher education, academic institutions based in non-English speaking countries are increasingly urged to produce contents in English to address international prospective students and personnel, as well as to increase their attractiveness. The demand for English translations in the institutional academic domain is consequently increasing at a rate exceeding the capacity of the translation profession. Resources for assisting non-native authors and translators in the production of appropriate texts in L2 are therefore required in order to help academic institutions and professionals streamline their translation workload. Some of these resources include: (i) parallel corpora to train machine translation systems and multilingual authoring tools; and (ii) translation memories for computer-aided tools. The purpose of this study is to create and evaluate reference resources like the ones mentioned in (i) and (ii) through the automatic sentence alignment of a large set of Italian and English as a Lingua Franca (ELF) institutional academic texts given as equivalent but not necessarily parallel (i.e. translated). In this framework, a set of aligning algorithms and alignment tools is examined in order to identify the most profitable one(s) in terms of accuracy and time- and cost-effectiveness. In order to determine the text pairs to align, a sample is selected according to document length similarity (characters) and subsequently evaluated in terms of extent of noisiness/parallelism, alignment accuracy and content leverageability. The results of these analyses serve as the basis for the creation of an aligned bilingual corpus of academic course descriptions, which is eventually used to create a translation memory in TMX format.

Involvement and detachment in David Cameron and Ed Miliband’s political discourse. An analysis of deictic shields on the pronoun ‘I’ in UK 2015 General Election news interviews

The goal of my study is to investigate the relationship between selected deictic shields on the pronoun ‘I’ and the involvement/detachment dichotomy in a sample of television news interviews. I focus on the use of personal pronouns in political discourse. Drawing upon Caffi’s (2007) classification of mitigating devices into bushes, hedges and shields, I focus on deictic shields on the pronoun ‘I’: I examine the way a selection of ‘I’-related deictic shields is employed in a collection of news interviews broadcast during the electoral campaign prior to the UK 2015 General Election. My purpose is to uncover the frequencies of each of the linguistic items selected and the pragmatic functions of those linguistic items in the involvement/detachment dichotomy. The research is structured as follows. Chapter 1 provides an account of previous studies on the three main areas of research: speech event analysis, institutional interaction and the news interview, and the UK 2015 General Election television programmes. Chapter 2 is centred on the involvement/detachment dichotomy: I provide an overview of nonlinguistic and linguistic features of involvement and detachment at all levels of sentence structure. Chapter 3 contains a detailed account of the data collection and data analysis process. Chapter 4 provides an accurate description of results in three steps: quantitative analysis, qualitative analysis and discussion of the pragmatic functions of the selected linguistic features of involvement and detachment. Chapter 5 includes a brief summary of the investigation, reviews the main findings, and indicates limitations of the study and possible inputs for further research. The results of the analysis confirm that, while some of the linguistic items examined point toward involvement, others have a detaching effect. I therefore conclude that deictic shields on the pronoun ‘I’ permit the realisation of the involvement/detachment dichotomy in the speech genre of the news interview.

Tree regularized neural networks for key point analysis

This thesis contributes to the ArgMining 2021 shared task on Key Point Analysis. Key Point Analysis entails extracting and calculating the prevalence of a concise list of the most prominent talking points, from an input corpus. These talking points are usually referred to as key points. Key point analysis is divided into two subtasks: Key Point Matching, which involves assigning a matching score to each key point/argument pair, and Key Point Generation, which consists of the generation of key points. The task of Key Point Matching was approached using different models: a pretrained Sentence Transformers model and a tree-constrained Graph Neural Network were tested. The best model was the fine-tuned Sentence Transformers, which achieved a mean Average Precision score of 0.75, ranking 12 compared to other participating teams. The model was then used for the subtask of Key Point Generation using the extractive method in the selection of key point candidates and the model developed for the previous subtask to evaluate them.

SynBA: A contextualized Synonim-Based adversarial Attack for text classification

With the advent of high-performance computing devices, deep neural networks have gained a lot of popularity in solving many Natural Language Processing tasks. However, they are also vulnerable to adversarial attacks, which are able to modify the input text in order to mislead the target model. Adversarial attacks are a serious threat to the security of deep neural networks, and they can be used to craft adversarial examples that steer the model towards a wrong decision. In this dissertation, we propose SynBA, a novel contextualized synonym-based adversarial attack for text classification. SynBA is based on the idea of replacing words in the input text with their synonyms, which are selected according to the context of the sentence. We show that SynBA successfully generates adversarial examples that are able to fool the target model with a high success rate. We demonstrate three advantages of this proposed approach: (1) effective - it outperforms state-of-the-art attacks by semantic similarity and perturbation rate, (2) utility-preserving - it preserves semantic content, grammaticality, and correct types classified by humans, and (3) efficient - it performs attacks faster than other methods.

Sviluppo di Retrieval-based Chatbot per l'Italiano con Transformer

L’Intelligenza Artificiale negli ultimi anni sta plasmando il futuro dell’umanità in quasi tutti i settori. È già il motore principale di diverse tecnologie emergenti come i big data, la robotica e l’IoT e continuerà ad agire come innovatore tecnologico nel futuro prossimo. Le recenti scoperte e migliorie sia nel campo dell’hardware che in quello matematico hanno migliorato l’efficienza e ridotto i tempi di esecuzione dei software. È in questo contesto che sta evolvendo anche il Natural Language Processing (NLP), un ramo dell’Intelligenza Artificiale che studia il modo in cui fornire ai computer l'abilità di comprendere un testo scritto o parlato allo stesso modo in cui lo farebbe un essere umano. Le ambiguità che distinguono la lingua naturale dalle altre rendono ardui gli studi in questo settore. Molti dei recenti sviluppi algoritmici su NLP si basano su tecnologie inventate decenni fa. La ricerca in questo settore è quindi in continua evoluzione. Questa tesi si pone l'obiettivo di sviluppare la logica di una chatbot help-desk per un'azienda privata. Lo scopo è, sottoposta una domanda da parte di un utente, restituire la risposta associata presente in una collezione domande-risposte. Il problema che questa tesi affronta è sviluppare un modello di NLP in grado di comprendere il significato semantico delle domande in input, poiché esse possono essere formulate in molteplici modi, preservando il contenuto semantico a discapito della sintassi. A causa delle ridotte dimensioni del dataset italiano proprietario su cui testare il modello chatbot, sono state eseguite molteplici sperimentazioni su un ulteriore dataset italiano con task affine. Attraverso diversi approcci di addestramento, tra cui apprendimento metrico, sono state raggiunte alte accuratezze sulle più comuni metriche di valutazione, confermando le capacità del modello proposto e sviluppato.

Hierarchical Multi-Label Text Classification in a Low-Resource Setting

In this thesis we address a multi-label hierarchical text classification problem in a low-resource setting and explore different approaches to identify the best one for our case. The goal is to train a model that classifies English school exercises according to a hierarchical taxonomy with few labeled data. The experiments made in this work employ different machine learning models and text representation techniques: CatBoost with tf-idf features, classifiers based on pre-trained models (mBERT, LASER), and SetFit, a framework for few-shot text classification. SetFit proved to be the most promising approach, achieving better performance when during training only a few labeled examples per class are available. However, this thesis does not consider all the hierarchical taxonomy, but only the first two levels: to address classification with the classes at the third level further experiments should be carried out, exploring methods for zero-shot text classification, data augmentation, and strategies to exploit the hierarchical structure of the taxonomy during training.