Quality and Aspect based Argument Generation

Natural Language Processing has always been one of the most popular topics in Artificial Intelligence. Argument-related research in NLP, such as argument detection, argument mining and argument generation, has been popular, especially in recent years. In our daily lives, we use arguments to express ourselves. The quality of arguments heavily impacts the effectiveness of our communications with others. In professional fields, such as legislation and academic areas, arguments of good quality play an even more critical role. Therefore, argument generation with good quality is a challenging research task that is also of great importance in NLP. The aim of this work is to investigate the automatic generation of arguments with good quality, according to the given topic, stance and aspect (control codes). To achieve this goal, a module based on BERT [17] which could judge an argument's quality is constructed. This module is used to assess the quality of the generated arguments. Another module based on GPT-2 [19] is implemented to generate arguments. Stances and aspects are also used as guidance when generating arguments. After combining all these models and techniques, the ranks of the generated arguments could be acquired to evaluate the final performance. This dissertation describes the architecture and experimental setup, analyzes the results of our experimentation, and discusses future directions.

SmartPantry - La dispensa intelligente

SmartPantry `e un applicazione per Android che si pone come obiettivo quello di rendere semplice e pratica la gestione virtuale delle dispense degli utenti. Oltre a questo implementa un recommender system dedicato al suggerimento di ricette adatte ai prodotti contenuti nella dispensa, per farlo l’algoritmo si avvale della distanza di Damerau-Levenshtein per eseguire Natural Language Processing in modo tale da interpretare gli ingredienti delle dispense degli utenti e poterli mappare ad una collezione di ingredienti mantenuti in un database remoto. All’interno di questo elaborato andremo ad analizzare i dettagli di progetta�zione ed implementativi di SmartPantry e degli algoritmi che la sostengono ponendo particolare attenzione agli aspetti qualitativi degli algoritmi di NLP e raccomandazione raccogliendo dati sufficienti a trarre conclusioni oggettive sulla precisione ed efficacia dei suddetti. Nell’ultimo capitolo vedremo come nonostante la presenza di margini di miglioramento, come versione 1.0, gli algoritmi abbiano restituito dei risultati pi`u che discreti

CartBot: un’applicazione di chat bot mobile per l’e-grocery

Sempre più negli ultimi anni si interagisce con i chatbot, software che simulano una conversazione con un essere umano utilizzando il linguaggio naturale. L’elaborato di tesi mira ad uno studio più approfondito della tematica, a partire da come tale tecnologia si è evoluta nel corso degli anni. Si procede analizzando le principali applicazioni dei bot, soffermandosi anche sui cambiamenti apportati dalla pandemia di Covid-19, ed evidenziando le principali ragioni che portano aziende e singoli al loro utilizzo. Inoltre, vengono descritti i diversi tipi di bot esistenti e viene analizzato il Natural Language Processing, ramo dell’Intelligenza Artificiale che mira alla comprensione del linguaggio naturale. Nei capitoli successivi viene descritto il progetto CartBot, un’applicazione di chat mobile per l’e-grocery, implementata come un chatbot che guida il cliente all’acquisto della spesa online. Vengono descritte le tecnologie utilizzate, con particolare riferimento al software di Google Dialogflow, che permette di sviluppare bot; inoltre viene analizzata come è stata effettuata la progettazione, sia lato front-end che back-end, allegando il flowchart, un diagramma di flusso realizzato per definire la sequenza di azioni e passaggi richiesti dal bot per effettuare l’acquisto. Infine, sono descritte le varie sottosezioni di CartBot, che riguardano la visualizzazione dei prodotti e il completamento dell’ordine, allegando screenshot dell’interfaccia finale ottenuta e inserendo il codice di alcune funzioni rilevanti.

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.

Implementazione di un sistema di traduzione automatica: un caso di studio per la traduzione italiano-cinese

Il lavoro di tesi presentato è nato da una collaborazione con il Politecnico di Macao, i referenti sono: Prof. Rita Tse, Prof. Marcus Im e Prof. Su-Kit Tang. L'obiettivo consiste nella creazione di un modello di traduzione automatica italiano-cinese e nell'osservarne il comportamento, al fine di determinare se sia o meno possibile l'impresa. Il trattato approfondisce l'argomento noto come Neural Language Processing (NLP), rientrando dunque nell'ambito delle traduzioni automatiche. Sono servizi che, attraverso l'ausilio dell'intelligenza artificiale sono in grado di elaborare il linguaggio naturale, per poi interpretarlo e tradurlo. NLP è una branca dell'informatica che unisce: computer science, intelligenza artificiale e studio di lingue. Dal punto di vista della ricerca, le più grandi sfide in questo ambito coinvolgono: il riconoscimento vocale (speech-recognition), comprensione del testo (natural-language understanding) e infine la generazione automatica di testo (natural-language generation). Lo stato dell'arte attuale è stato definito dall'articolo "Attention is all you need" \cite{vaswani2017attention}, presentato nel 2017 a partire da una collaborazione di ricercatori della Cornell University.\\ I modelli di traduzione automatica più noti ed utilizzati al momento sono i Neural Machine Translators (NMT), ovvero modelli che attraverso le reti neurali artificiali profonde, sono in grado effettuare traduzioni o predizioni. La qualità delle traduzioni è particolarmente buona, tanto da arrivare quasi a raggiungere la qualità di una traduzione umana. Il lavoro infatti si concentrerà largamente sullo studio e utilizzo di NMT, allo scopo di proporre un modello funzionale e che sia in grado di performare al meglio nelle traduzioni da italiano a cinese e viceversa.

Sviluppo di Metodi di Soft Labeling per la Multi-Document Summarization in Ambito Legale

In questo elaborato viene trattata l’analisi del problema di soft labeling applicato alla multi-document summarization, in particolare vengono testate varie tecniche per estrarre frasi rilevanti dai documenti presi in dettaglio, al fine di fornire al modello di summarization quelle di maggior rilievo e più informative per il riassunto da generare. Questo problema nasce per far fronte ai limiti che presentano i modelli di summarization attualmente a disposizione, che possono processare un numero limitato di frasi; sorge quindi la necessità di filtrare le informazioni più rilevanti quando il lavoro si applica a documenti lunghi. Al fine di scandire la metrica di importanza, vengono presi come riferimento metodi sintattici, semantici e basati su rappresentazione a grafi AMR. Il dataset preso come riferimento è Multi-LexSum, che include tre granularità di summarization di testi legali. L’analisi in questione si compone quindi della fase di estrazione delle frasi dai documenti, della misurazione delle metriche stabilite e del passaggio al modello stato dell’arte PRIMERA per l’elaborazione del riassunto. Il testo ottenuto viene poi confrontato con il riassunto target già fornito, considerato come ottimale; lavorando in queste condizioni l’obiettivo è di definire soglie ottimali di upper-bound per l’accuratezza delle metriche, che potrebbero ampliare il lavoro ad analisi più dettagliate qualora queste superino lo stato dell’arte attuale.

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.

prism: rinfrangere distribuzioni di probabilità in riassunti astrattivi mediante strategie di decoding

Negli ultimi quattro anni la summarization astrattiva è stata protagonista di una evoluzione senza precedenti dettata da nuovi language model neurali, architetture transformer-based, elevati spazi dimensionali, ampi dataset e innovativi task di pre-training. In questo contesto, le strategie di decoding convertono le distribuzioni di probabilità predette da un modello in un testo artificiale, il quale viene composto in modo auto regressivo. Nonostante il loro cruciale impatto sulla qualità dei riassunti inferiti, il ruolo delle strategie di decoding è frequentemente trascurato e sottovalutato. Di fronte all'elevato numero di tecniche e iperparametri, i ricercatori necessitano di operare scelte consapevoli per ottenere risultati più affini agli obiettivi di generazione. Questa tesi propone il primo studio altamente comprensivo sull'efficacia ed efficienza delle strategie di decoding in task di short, long e multi-document abstractive summarization. Diversamente dalle pubblicazioni disponibili in letteratura, la valutazione quantitativa comprende 5 metriche automatiche, analisi temporali e carbon footprint. I risultati ottenuti dimostrano come non vi sia una strategia di decoding dominante, ma come ciascuna possieda delle caratteristiche adatte a task e dataset specifici. I contributi proposti hanno l'obiettivo di neutralizzare il gap di conoscenza attuale e stimolare lo sviluppo di nuove tecniche di decoding.

Rappresentazione semantica delle informazioni in ambito storico e giuridico

Rappresentazione della conoscenza in banca di dati testuali non strutturati in lingua Italiana.

Antibody-Antigen Interface Characterization with Contextualized Pre-trained Embeddings

Artificial Intelligence (AI) has substantially influenced numerous disciplines in recent years. Biology, chemistry, and bioinformatics are among them, with significant advances in protein structure prediction, paratope prediction, protein-protein interactions (PPIs), and antibody-antigen interactions. Understanding PPIs is critical since they are responsible for practically everything living and have several uses in vaccines, cancer, immunology, and inflammatory illnesses. Machine Learning (ML) offers enormous potential for effectively simulating antibody-antigen interactions and improving in-silico optimization of therapeutic antibodies for desired features, including binding activity, stability, and low immunogenicity. This research looks at the use of AI algorithms to better understand antibody-antigen interactions, and it further expands and explains several difficulties encountered in the field. Furthermore, we contribute by presenting a method that outperforms existing state-of-the-art strategies in paratope prediction from sequence data.

Design And Development Of A Named Entity Based Question Answering System For Malayalam Language

This is a Named Entity Based Question Answering System for Malayalam Language. Although a vast amount of information is available today in digital form, no effective information access mechanism exists to provide humans with convenient information access. Information Retrieval and Question Answering systems are the two mechanisms available now for information access. Information systems typically return a long list of documents in response to a user’s query which are to be skimmed by the user to determine whether they contain an answer. But a Question Answering System allows the user to state his/her information need as a natural language question and receives most appropriate answer in a word or a sentence or a paragraph. This system is based on Named Entity Tagging and Question Classification. Document tagging extracts useful information from the documents which will be used in finding the answer to the question. Question Classification extracts useful information from the question to determine the type of the question and the way in which the question is to be answered. Various Machine Learning methods are used to tag the documents. Rule-Based Approach is used for Question Classification. Malayalam belongs to the Dravidian family of languages and is one of the four major languages of this family. It is one of the 22 Scheduled Languages of India with official language status in the state of Kerala. It is spoken by 40 million people. Malayalam is a morphologically rich agglutinative language and relatively of free word order. Also Malayalam has a productive morphology that allows the creation of complex words which are often highly ambiguous. Document tagging tools such as Parts-of-Speech Tagger, Phrase Chunker, Named Entity Tagger, and Compound Word Splitter are developed as a part of this research work. No such tools were available for Malayalam language. Finite State Transducer, High Order Conditional Random Field, Artificial Immunity System Principles, and Support Vector Machines are the techniques used for the design of these document preprocessing tools. This research work describes how the Named Entity is used to represent the documents. Single sentence questions are used to test the system. Overall Precision and Recall obtained are 88.5% and 85.9% respectively. This work can be extended in several directions. The coverage of non-factoid questions can be increased and also it can be extended to include open domain applications. Reference Resolution and Word Sense Disambiguation techniques are suggested as the future enhancements

OntoTag - A Linguistic and Ontological Annotation Model Suitable for the Semantic Web

OntoTag - A Linguistic and Ontological Annotation Model Suitable for the Semantic Web 1. INTRODUCTION. LINGUISTIC TOOLS AND ANNOTATIONS: THEIR LIGHTS AND SHADOWS Computational Linguistics is already a consolidated research area. It builds upon the results of other two major ones, namely Linguistics and Computer Science and Engineering, and it aims at developing computational models of human language (or natural language, as it is termed in this area). Possibly, its most well-known applications are the different tools developed so far for processing human language, such as machine translation systems and speech recognizers or dictation programs. These tools for processing human language are commonly referred to as linguistic tools. Apart from the examples mentioned above, there are also other types of linguistic tools that perhaps are not so well-known, but on which most of the other applications of Computational Linguistics are built. These other types of linguistic tools comprise POS taggers, natural language parsers and semantic taggers, amongst others. All of them can be termed linguistic annotation tools. Linguistic annotation tools are important assets. In fact, POS and semantic taggers (and, to a lesser extent, also natural language parsers) have become critical resources for the computer applications that process natural language. Hence, any computer application that has to analyse a text automatically and ‘intelligently’ will include at least a module for POS tagging. The more an application needs to ‘understand’ the meaning of the text it processes, the more linguistic tools and/or modules it will incorporate and integrate. However, linguistic annotation tools have still some limitations, which can be summarised as follows: 1. Normally, they perform annotations only at a certain linguistic level (that is, Morphology, Syntax, Semantics, etc.). 2. They usually introduce a certain rate of errors and ambiguities when tagging. This error rate ranges from 10 percent up to 50 percent of the units annotated for unrestricted, general texts. 3. Their annotations are most frequently formulated in terms of an annotation schema designed and implemented ad hoc. A priori, it seems that the interoperation and the integration of several linguistic tools into an appropriate software architecture could most likely solve the limitations stated in (1). Besides, integrating several linguistic annotation tools and making them interoperate could also minimise the limitation stated in (2). Nevertheless, in the latter case, all these tools should produce annotations for a common level, which would have to be combined in order to correct their corresponding errors and inaccuracies. Yet, the limitation stated in (3) prevents both types of integration and interoperation from being easily achieved. In addition, most high-level annotation tools rely on other lower-level annotation tools and their outputs to generate their own ones. For example, sense-tagging tools (operating at the semantic level) often use POS taggers (operating at a lower level, i.e., the morphosyntactic) to identify the grammatical category of the word or lexical unit they are annotating. Accordingly, if a faulty or inaccurate low-level annotation tool is to be used by other higher-level one in its process, the errors and inaccuracies of the former should be minimised in advance. Otherwise, these errors and inaccuracies would be transferred to (and even magnified in) the annotations of the high-level annotation tool. Therefore, it would be quite useful to find a way to (i) correct or, at least, reduce the errors and the inaccuracies of lower-level linguistic tools; (ii) unify the annotation schemas of different linguistic annotation tools or, more generally speaking, make these tools (as well as their annotations) interoperate. Clearly, solving (i) and (ii) should ease the automatic annotation of web pages by means of linguistic tools, and their transformation into Semantic Web pages (Berners-Lee, Hendler and Lassila, 2001). Yet, as stated above, (ii) is a type of interoperability problem. There again, ontologies (Gruber, 1993; Borst, 1997) have been successfully applied thus far to solve several interoperability problems. Hence, ontologies should help solve also the problems and limitations of linguistic annotation tools aforementioned. Thus, to summarise, the main aim of the present work was to combine somehow these separated approaches, mechanisms and tools for annotation from Linguistics and Ontological Engineering (and the Semantic Web) in a sort of hybrid (linguistic and ontological) annotation model, suitable for both areas. This hybrid (semantic) annotation model should (a) benefit from the advances, models, techniques, mechanisms and tools of these two areas; (b) minimise (and even solve, when possible) some of the problems found in each of them; and (c) be suitable for the Semantic Web. The concrete goals that helped attain this aim are presented in the following section. 2. GOALS OF THE PRESENT WORK As mentioned above, the main goal of this work was to specify a hybrid (that is, linguistically-motivated and ontology-based) model of annotation suitable for the Semantic Web (i.e. it had to produce a semantic annotation of web page contents). This entailed that the tags included in the annotations of the model had to (1) represent linguistic concepts (or linguistic categories, as they are termed in ISO/DCR (2008)), in order for this model to be linguistically-motivated; (2) be ontological terms (i.e., use an ontological vocabulary), in order for the model to be ontology-based; and (3) be structured (linked) as a collection of ontology-based triples, as in the usual Semantic Web languages (namely RDF(S) and OWL), in order for the model to be considered suitable for the Semantic Web. Besides, to be useful for the Semantic Web, this model should provide a way to automate the annotation of web pages. As for the present work, this requirement involved reusing the linguistic annotation tools purchased by the OEG research group (http://www.oeg-upm.net), but solving beforehand (or, at least, minimising) some of their limitations. Therefore, this model had to minimise these limitations by means of the integration of several linguistic annotation tools into a common architecture. Since this integration required the interoperation of tools and their annotations, ontologies were proposed as the main technological component to make them effectively interoperate. From the very beginning, it seemed that the formalisation of the elements and the knowledge underlying linguistic annotations within an appropriate set of ontologies would be a great step forward towards the formulation of such a model (henceforth referred to as OntoTag). Obviously, first, to combine the results of the linguistic annotation tools that operated at the same level, their annotation schemas had to be unified (or, preferably, standardised) in advance. This entailed the unification (id. standardisation) of their tags (both their representation and their meaning), and their format or syntax. Second, to merge the results of the linguistic annotation tools operating at different levels, their respective annotation schemas had to be (a) made interoperable and (b) integrated. And third, in order for the resulting annotations to suit the Semantic Web, they had to be specified by means of an ontology-based vocabulary, and structured by means of ontology-based triples, as hinted above. Therefore, a new annotation scheme had to be devised, based both on ontologies and on this type of triples, which allowed for the combination and the integration of the annotations of any set of linguistic annotation tools. This annotation scheme was considered a fundamental part of the model proposed here, and its development was, accordingly, another major objective of the present work. All these goals, aims and objectives could be re-stated more clearly as follows: Goal 1: Development of a set of ontologies for the formalisation of the linguistic knowledge relating linguistic annotation. Sub-goal 1.1: Ontological formalisation of the EAGLES (1996a; 1996b) de facto standards for morphosyntactic and syntactic annotation, in a way that helps respect the triple structure recommended for annotations in these works (which is isomorphic to the triple structures used in the context of the Semantic Web). Sub-goal 1.2: Incorporation into this preliminary ontological formalisation of other existing standards and standard proposals relating the levels mentioned above, such as those currently under development within ISO/TC 37 (the ISO Technical Committee dealing with Terminology, which deals also with linguistic resources and annotations). Sub-goal 1.3: Generalisation and extension of the recommendations in EAGLES (1996a; 1996b) and ISO/TC 37 to the semantic level, for which no ISO/TC 37 standards have been developed yet. Sub-goal 1.4: Ontological formalisation of the generalisations and/or extensions obtained in the previous sub-goal as generalisations and/or extensions of the corresponding ontology (or ontologies). Sub-goal 1.5: Ontological formalisation of the knowledge required to link, combine and unite the knowledge represented in the previously developed ontology (or ontologies). Goal 2: Development of OntoTag’s annotation scheme, a standard-based abstract scheme for the hybrid (linguistically-motivated and ontological-based) annotation of texts. Sub-goal 2.1: Development of the standard-based morphosyntactic annotation level of OntoTag’s scheme. This level should include, and possibly extend, the recommendations of EAGLES (1996a) and also the recommendations included in the ISO/MAF (2008) standard draft. Sub-goal 2.2: Development of the standard-based syntactic annotation level of the hybrid abstract scheme. This level should include, and possibly extend, the recommendations of EAGLES (1996b) and the ISO/SynAF (2010) standard draft. Sub-goal 2.3: Development of the standard-based semantic annotation level of OntoTag’s (abstract) scheme. Sub-goal 2.4: Development of the mechanisms for a convenient integration of the three annotation levels already mentioned. These mechanisms should take into account the recommendations included in the ISO/LAF (2009) standard draft. Goal 3: Design of OntoTag’s (abstract) annotation architecture, an abstract architecture for the hybrid (semantic) annotation of texts (i) that facilitates the integration and interoperation of different linguistic annotation tools, and (ii) whose results comply with OntoTag’s annotation scheme. Sub-goal 3.1: Specification of the decanting processes that allow for the classification and separation, according to their corresponding levels, of the results of the linguistic tools annotating at several different levels. Sub-goal 3.2: Specification of the standardisation processes that allow (a) complying with the standardisation requirements of OntoTag’s annotation scheme, as well as (b) combining the results of those linguistic tools that share some level of annotation. Sub-goal 3.3: Specification of the merging processes that allow for the combination of the output annotations and the interoperation of those linguistic tools that share some level of annotation. Sub-goal 3.4: Specification of the merge processes that allow for the integration of the results and the interoperation of those tools performing their annotations at different levels. Goal 4: Generation of OntoTagger’s schema, a concrete instance of OntoTag’s abstract scheme for a concrete set of linguistic annotations. These linguistic annotations result from the tools and the resources available in the research group, namely • Bitext’s DataLexica (http://www.bitext.com/EN/datalexica.asp), • LACELL’s (POS) tagger (http://www.um.es/grupos/grupo-lacell/quees.php), • Connexor’s FDG (http://www.connexor.eu/technology/machinese/glossary/fdg/), and • EuroWordNet (Vossen et al., 1998). This schema should help evaluate OntoTag’s underlying hypotheses, stated below. Consequently, it should implement, at least, those levels of the abstract scheme dealing with the annotations of the set of tools considered in this implementation. This includes the morphosyntactic, the syntactic and the semantic levels. Goal 5: Implementation of OntoTagger’s configuration, a concrete instance of OntoTag’s abstract architecture for this set of linguistic tools and annotations. This configuration (1) had to use the schema generated in the previous goal; and (2) should help support or refute the hypotheses of this work as well (see the next section). Sub-goal 5.1: Implementation of the decanting processes that facilitate the classification and separation of the results of those linguistic resources that provide annotations at several different levels (on the one hand, LACELL’s tagger operates at the morphosyntactic level and, minimally, also at the semantic level; on the other hand, FDG operates at the morphosyntactic and the syntactic levels and, minimally, at the semantic level as well). Sub-goal 5.2: Implementation of the standardisation processes that allow (i) specifying the results of those linguistic tools that share some level of annotation according to the requirements of OntoTagger’s schema, as well as (ii) combining these shared level results. In particular, all the tools selected perform morphosyntactic annotations and they had to be conveniently combined by means of these processes. Sub-goal 5.3: Implementation of the merging processes that allow for the combination (and possibly the improvement) of the annotations and the interoperation of the tools that share some level of annotation (in particular, those relating the morphosyntactic level, as in the previous sub-goal). Sub-goal 5.4: Implementation of the merging processes that allow for the integration of the different standardised and combined annotations aforementioned, relating all the levels considered. Sub-goal 5.5: Improvement of the semantic level of this configuration by adding a named entity recognition, (sub-)classification and annotation subsystem, which also uses the named entities annotated to populate a domain ontology, in order to provide a concrete application of the present work in the two areas involved (the Semantic Web and Corpus Linguistics). 3. MAIN RESULTS: ASSESSMENT OF ONTOTAG’S UNDERLYING HYPOTHESES The model developed in the present thesis tries to shed some light on (i) whether linguistic annotation tools can effectively interoperate; (ii) whether their results can be combined and integrated; and, if they can, (iii) how they can, respectively, interoperate and be combined and integrated. Accordingly, several hypotheses had to be supported (or rejected) by the development of the OntoTag model and OntoTagger (its implementation). The hypotheses underlying OntoTag are surveyed below. Only one of the hypotheses (H.6) was rejected; the other five could be confirmed. H.1 The annotations of different levels (or layers) can be integrated into a sort of overall, comprehensive, multilayer and multilevel annotation, so that their elements can complement and refer to each other. • CONFIRMED by the development of: o OntoTag’s annotation scheme, o OntoTag’s annotation architecture, o OntoTagger’s (XML, RDF, OWL) annotation schemas, o OntoTagger’s configuration. H.2 Tool-dependent annotations can be mapped onto a sort of tool-independent annotations and, thus, can be standardised. • CONFIRMED by means of the standardisation phase incorporated into OntoTag and OntoTagger for the annotations yielded by the tools. H.3 Standardisation should ease: H.3.1: The interoperation of linguistic tools. H.3.2: The comparison, combination (at the same level and layer) and integration (at different levels or layers) of annotations. • H.3 was CONFIRMED by means of the development of OntoTagger’s ontology-based configuration: o Interoperation, comparison, combination and integration of the annotations of three different linguistic tools (Connexor’s FDG, Bitext’s DataLexica and LACELL’s tagger); o Integration of EuroWordNet-based, domain-ontology-based and named entity annotations at the semantic level. o Integration of morphosyntactic, syntactic and semantic annotations. H.4 Ontologies and Semantic Web technologies (can) play a crucial role in the standardisation of linguistic annotations, by providing consensual vocabularies and standardised formats for annotation (e.g., RDF triples). • CONFIRMED by means of the development of OntoTagger’s RDF-triple-based annotation schemas. H.5 The rate of errors introduced by a linguistic tool at a given level, when annotating, can be reduced automatically by contrasting and combining its results with the ones coming from other tools, operating at the same level. However, these other tools might be built following a different technological (stochastic vs. rule-based, for example) or theoretical (dependency vs. HPS-grammar-based, for instance) approach. • CONFIRMED by the results yielded by the evaluation of OntoTagger. H.6 Each linguistic level can be managed and annotated independently. • REJECTED: OntoTagger’s experiments and the dependencies observed among the morphosyntactic annotations, and between them and the syntactic annotations. In fact, Hypothesis H.6 was already rejected when OntoTag’s ontologies were developed. We observed then that several linguistic units stand on an interface between levels, belonging thereby to both of them (such as morphosyntactic units, which belong to both the morphological level and the syntactic level). Therefore, the annotations of these levels overlap and cannot be handled independently when merged into a unique multileveled annotation. 4. OTHER MAIN RESULTS AND CONTRIBUTIONS First, interoperability is a hot topic for both the linguistic annotation community and the whole Computer Science field. The specification (and implementation) of OntoTag’s architecture for the combination and integration of linguistic (annotation) tools and annotations by means of ontologies shows a way to make these different linguistic annotation tools and annotations interoperate in practice. Second, as mentioned above, the elements involved in linguistic annotation were formalised in a set (or network) of ontologies (OntoTag’s linguistic ontologies). • On the one hand, OntoTag’s network of ontologies consists of − The Linguistic Unit Ontology (LUO), which includes a mostly hierarchical formalisation of the different types of linguistic elements (i.e., units) identifiable in a written text; − The Linguistic Attribute Ontology (LAO), which includes also a mostly hierarchical formalisation of the different types of features that characterise the linguistic units included in the LUO; − The Linguistic Value Ontology (LVO), which includes the corresponding formalisation of the different values that the attributes in the LAO can take; − The OIO (OntoTag’s Integration Ontology), which  Includes the knowledge required to link, combine and unite the knowledge represented in the LUO, the LAO and the LVO;  Can be viewed as a knowledge representation ontology that describes the most elementary vocabulary used in the area of annotation. • On the other hand, OntoTag’s ontologies incorporate the knowledge included in the different standards and recommendations for linguistic annotation released so far, such as those developed within the EAGLES and the SIMPLE European projects or by the ISO/TC 37 committee: − As far as morphosyntactic annotations are concerned, OntoTag’s ontologies formalise the terms in the EAGLES (1996a) recommendations and their corresponding terms within the ISO Morphosyntactic Annotation Framework (ISO/MAF, 2008) standard; − As for syntactic annotations, OntoTag’s ontologies incorporate the terms in the EAGLES (1996b) recommendations and their corresponding terms within the ISO Syntactic Annotation Framework (ISO/SynAF, 2010) standard draft; − Regarding semantic annotations, OntoTag’s ontologies generalise and extend the recommendations in EAGLES (1996a; 1996b) and, since no stable standards or standard drafts have been released for semantic annotation by ISO/TC 37 yet, they incorporate the terms in SIMPLE (2000) instead; − The terms coming from all these recommendations and standards were supplemented by those within the ISO Data Category Registry (ISO/DCR, 2008) and also of the ISO Linguistic Annotation Framework (ISO/LAF, 2009) standard draft when developing OntoTag’s ontologies. Third, we showed that the combination of the results of tools annotating at the same level can yield better results (both in precision and in recall) than each tool separately. In particular, 1. OntoTagger clearly outperformed two of the tools integrated into its configuration, namely DataLexica and FDG in all the combination sub-phases in which they overlapped (i.e. POS tagging, lemma annotation and morphological feature annotation). As far as the remaining tool is concerned, i.e. LACELL’s tagger, it was also outperformed by OntoTagger in POS tagging and lemma annotation, and it did not behave better than OntoTagger in the morphological feature annotation layer. 2. As an immediate result, this implies that a) This type of combination architecture configurations can be applied in order to improve significantly the accuracy of linguistic annotations; and b) Concerning the morphosyntactic level, this could be regarded as a way of constructing more robust and more accurate POS tagging systems. Fourth, Semantic Web annotations are usually performed by humans or else by machine learning systems. Both of them leave much to be desired: the former, with respect to their annotation rate; the latter, with respect to their (average) precision and recall. In this work, we showed how linguistic tools can be wrapped in order to annotate automatically Semantic Web pages using ontologies. This entails their fast, robust and accurate semantic annotation. As a way of example, as mentioned in Sub-goal 5.5, we developed a particular OntoTagger module for the recognition, classification and labelling of named entities, according to the MUC and ACE tagsets (Chinchor, 1997; Doddington et al., 2004). These tagsets were further specified by means of a domain ontology, namely the Cinema Named Entities Ontology (CNEO). This module was applied to the automatic annotation of ten different web pages containing cinema reviews (that is, around 5000 words). In addition, the named entities annotated with this module were also labelled as instances (or individuals) of the classes included in the CNEO and, then, were used to populate this domain ontology. • The statistical results obtained from the evaluation of this particular module of OntoTagger can be summarised as follows. On the one hand, as far as recall (R) is concerned, (R.1) the lowest value was 76,40% (for file 7); (R.2) the highest value was 97, 50% (for file 3); and (R.3) the average value was 88,73%. On the other hand, as far as the precision rate (P) is concerned, (P.1) its minimum was 93,75% (for file 4); (R.2) its maximum was 100% (for files 1, 5, 7, 8, 9, and 10); and (R.3) its average value was 98,99%. • These results, which apply to the tasks of named entity annotation and ontology population, are extraordinary good for both of them. They can be explained on the basis of the high accuracy of the annotations provided by OntoTagger at the lower levels (mainly at the morphosyntactic level). However, they should be conveniently qualified, since they might be too domain- and/or language-dependent. It should be further experimented how our approach works in a different domain or a different language, such as French, English, or German. • In any case, the results of this application of Human Language Technologies to Ontology Population (and, accordingly, to Ontological Engineering) seem very promising and encouraging in order for these two areas to collaborate and complement each other in the area of semantic annotation. Fifth, as shown in the State of the Art of this work, there are different approaches and models for the semantic annotation of texts, but all of them focus on a particular view of the semantic level. Clearly, all these approaches and models should be integrated in order to bear a coherent and joint semantic annotation level. OntoTag shows how (i) these semantic annotation layers could be integrated together; and (ii) they could be integrated with the annotations associated to other annotation levels. Sixth, we identified some recommendations, best practices and lessons learned for annotation standardisation, interoperation and merge. They show how standardisation (via ontologies, in this case) enables the combination, integration and interoperation of different linguistic tools and their annotations into a multilayered (or multileveled) linguistic annotation, which is one of the hot topics in the area of Linguistic Annotation. And last but not least, OntoTag’s annotation scheme and OntoTagger’s annotation schemas show a way to formalise and annotate coherently and uniformly the different units and features associated to the different levels and layers of linguistic annotation. This is a great scientific step ahead towards the global standardisation of this area, which is the aim of ISO/TC 37 (in particular, Subcommittee 4, dealing with the standardisation of linguistic annotations and resources).

Regional and hemispheric determinants of language laterality: Implications for preoperative fMRI.

Language is typically a function of the left hemisphere but the right hemisphere is also essential in some healthy individuals and patients. This inter-subject variability necessitates the localization of language function, at the individual level, prior to neurosurgical intervention. Such assessments are typically made by comparing left and right hemisphere language function to determine "language lateralization" using clinical tests or fMRI. Here, we show that language function needs to be assessed at the region and hemisphere specific level, because laterality measures can be misleading. Using fMRI data from 82 healthy participants, we investigated the degree to which activation for a semantic word matching task was lateralized in 50 different brain regions and across the entire cortex. This revealed two novel findings. First, the degree to which language is lateralized across brain regions and between subjects was primarily driven by differences in right hemisphere activation rather than differences in left hemisphere activation. Second, we found that healthy subjects who have relatively high left lateralization in the angular gyrus also have relatively low left lateralization in the ventral precentral gyrus. These findings illustrate spatial heterogeneity in language lateralization that is lost when global laterality measures are considered. It is likely that the complex spatial variability we observed in healthy controls is more exaggerated in patients with brain damage. We therefore highlight the importance of investigating within hemisphere regional variations in fMRI activation, prior to neuro-surgical intervention, to determine how each hemisphere and each region contributes to language processing. Hum Brain Mapp, 2010. © 2010 Wiley-Liss, Inc.

Reconciling phonological neighborhood effects in speech production through single trial analysis

A crucial step for understanding how lexical knowledge is represented is to describe the relative similarity of lexical items, and how it influences language processing. Previous studies of the effects of form similarity on word production have reported conflicting results, notably within and across languages. The aim of the present study was to clarify this empirical issue to provide specific constraints for theoretical models of language production. We investigated the role of phonological neighborhood density in a large-scale picture naming experiment using fine-grained statistical models. The results showed that increasing phonological neighborhood density has a detrimental effect on naming latencies, and re-analyses of independently obtained data sets provide supplementary evidence for this effect. Finally, we reviewed a large body of evidence concerning phonological neighborhood density effects in word production, and discussed the occurrence of facilitatory and inhibitory effects in accuracy measures. The overall pattern shows that phonological neighborhood generates two opposite forces, one facilitatory and one inhibitory. In cases where speech production is disrupted (e.g. certain aphasic symptoms), the facilitatory component may emerge, but inhibitory processes dominate in efficient naming by healthy speakers. These findings are difficult to accommodate in terms of monitoring processes, but can be explained within interactive activation accounts combining phonological facilitation and lexical competition.