Littérature scientifique sur le sujet « METRICA CORPOREA »

In this article we discuss several metrics of coherence defined using centering theory and investigate the usefulness of such metrics for information ordering in automatic text generation. We estimate empirically which is the most promising metric and how useful this metric is using a general methodology applied on several corpora. Our main result is that the simplest metric (which relies exclusively on NOCB transitions) sets a robust baseline that cannot be outperformed by other metrics which make use of additional centering-based features. This baseline can be used for the development of both text-to-text and concept-to-text generation systems.

Abstract The development and application of biological ontologies have increased significantly in recent years. These ontologies can be retrieved from different repositories, which do not provide much information about quality aspects of the ontologies. In the past years, some ontology structural metrics have been proposed, but their validity as measurement instrument has not been sufficiently studied to date. In this work, we evaluate a set of reproducible and objective ontology structural metrics. Given the lack of standard methods for this purpose, we have applied an evaluation method based on the stability and goodness of the classifications of ontologies produced by each metric on an ontology corpus. The evaluation has been done using ontology repositories as corpora. More concretely, we have used 119 ontologies from the OBO Foundry repository and 78 ontologies from AgroPortal. First, we study the correlations between the metrics. Second, we study whether the clusters for a given metric are stable and have a good structure. The results show that the existing correlations are not biasing the evaluation, there are no metrics generating unstable clusterings and all the metrics evaluated provide at least reasonable clustering structure. Furthermore, our work permits to review and suggest the most reliable ontology structural metrics in terms of stability and goodness of their classifications. Availability: http://sele.inf.um.es/ontology-metrics

Recent work on Arabic metrics aims at developing a coherent research programme which relies on the systematic analysis of electronic corpora. The computer program XALIYL performs, for any line of ancient Arabic poetry, an automatic recognition of the metre used. This operation takes place whatever the length of the verses, and regardless of whether they are encoded in ordinary Arabic script (with the addition of vowels) or by means of the TRS system, which relates functionally to ordinary Arabic script. XALIYL produces a textual database that contains the syllabic decomposition for each hemistich of each line, as well as its metrical analysis. It can cope not only with the general problems linked to re-syllabification and sandhi, but also with problems of syllabification specific to Arabic metrics. Errors due to the metrical scanning or to the editing of poems can be located automatically. Moreover, by allowing a computerised search for formulae, XALIYL provides significant information on the “formulaic systems” of ancient Arabic poetry.

AbstractSpelling errors in digital documents are often caused by operational and cognitive mistakes, or by the lack of full knowledge about the language of the written documents. Computer-assisted solutions can help to detect and suggest replacements. In this paper, we present a new string distance metric for the Persian language to rank respelling suggestions of a misspelled Persian word by considering the effects of keyboard layout on typographical spelling errors as well as the homomorphic and homophonic aspects of words for orthographical misspellings. We also consider the misspellings caused by disregarded diacritics. Since the proposed string distance metric is custom-designed for the Persian language, we present the spelling aspects of the Persian language such as homomorphs, homophones, and diacritics. We then present our statistical analysis of a set of large Persian corpora to identify the causes and the types of Persian spelling errors. We show that the proposed string distance metric has a higher mean average precision and a higher mean reciprocal rank in ranking respelling candidates of Persian misspellings in comparison with other metrics such as the Hamming, Levenshtein, Damerau–Levenshtein, Wagner–Fischer, and Jaro–Winkler metrics.

The corpus-based identification of those lexical units which serve to describe a given specialized domain usually becomes a complex task, where an analysis oriented to the frequency of words and the likelihood of lexical associations is often ineffective. The goal of this article is to demonstrate that a user-adjustable composite metric such as SRC can accommodate to the diversity of domain-specific glossaries to be constructed from small- and medium-sized specialized corpora of non-structured texts. Unlike for most of the research in automatic term extraction, where single metrics are usually combined indiscriminately to produce the best results, SRC is grounded on the theoretical principles of salience, relevance and cohesion, which have been rationally implemented in the three components of this metric.

In the field of metrics, it has long been observed that verse lines tend to be more regular or restricted towards the end (Arnold 1905). This has led to the Strict End Hypothesis [SEH], which proposes a general versification principle of universal scope (Hayes 1983). This paper argues that two main challenges hinder the substantiation of the SEH in a broad typological sample of unrelated verse corpora. First, the concept of strictness is too coarse and needs to be narrowed down to testable features or subcomponents. Second, explicit measures need to be developed which enable the systematic comparison of corpora, particularly when trying to capture potentially gradient features such as the relative faithfulness to a metrical template. This study showcases how to overcome these issues by analysing the entropy at different positions in the line for corpora in five languages (English, Dutch, Sanskrit, Estonian, Berber). Finally, I argue that, if the SEH is shown to be typologically robust, shared human cognitive features may provide a partial explanation for this puzzling asymmetry in verse lines.

AbstractUser simulation is an important research area in the field of spoken dialogue systems (SDSs) because collecting and annotating real human–machine interactions is often expensive and time-consuming. However, such data are generally required for designing, training and assessing dialogue systems. User simulations are especially needed when using machine learning methods for optimizing dialogue management strategies such as Reinforcement Learning, where the amount of data necessary for training is larger than existing corpora. The quality of the user simulation is therefore of crucial importance because it dramatically influences the results in terms of SDS performance analysis and the learnt strategy. Assessment of the quality of simulated dialogues and user simulation methods is an open issue and, although assessment metrics are required, there is no commonly adopted metric. In this paper, we give a survey of User Simulations Metrics in the literature, propose some extensions and discuss these metrics in terms of a list of desired features.

While a number of tools have been developed for researchers to compute the lexical characteristics of words, extant resources are limited in their useability and functionality. Specifically, some tools require users to have some prior knowledge of some aspects of the applications, and not all tools allow users to specify their own corpora. Additionally, current tools are also limited in terms of the range of metrics that they can compute. To address these methodological gaps, this article introduces LexiCAL, a fast, simple, and intuitive calculator for lexical variables. Specifically, LexiCAL is a standalone executable that provides options for users to calculate a range of theoretically influential surface, orthographic, phonological, and phonographic metrics for any alphabetic language, using any user-specified input, corpus file, and phonetic system. LexiCAL also comes with a set of well-documented Python scripts for each metric, that can be reproduced and/or modified for other research purposes.

Slavic languages are generally assumed to possess rich morphological features with free syntactic word order. Exploring this complexity trade-off can help us better understand the relationship between morphology and syntax within natural languages. However, few quantitative investigations have been carried out into this relationship within Slavic languages. Based on 34 annotated corpora from Universal Dependencies, this paper paid special attention to the correlations between morphology and syntax within Slavic languages by applying two metrics of morphological richness and two of word order freedom, respectively. Our findings are as follows. First, the quantitative metrics adopted can well capture the distributions of morphological richness and word order freedom of languages. Second, the metrics can corroborate the correlation between morphological richness and word order freedom. Within Slavic languages, this correlation is moderate and statistically significant. Precisely, the richer the morphology, the less strict the word order. Third, Slavic languages can be clustered into three subgroups based on classification models. Most importantly, ancient Slavic languages are characterized by richer morphology and more flexible word order than modern ones. Fourth, as two possible disturbing factors, corpus size does not greatly affect the results of the metrics, whereas corpus genre does play an important part in the measurements of word order freedom. Specifically, the word order of formal written genres tends to be more rigid than that of informal written and spoken ones. Overall, based on annotated corpora, the results verify the negative correlation between morphological richness and word order rigidity within Slavic languages, which might shed light on the dynamic relations between morphology and syntax of natural languages and provide quantitative instantiations of how languages encode lexical and syntactic information for the purpose of efficient communication.

Using part-of-speech (POS) tagged corpora, Hudson (1994) reports that approximately 37% of English tokens are nouns, where ‘noun’ is a superordinate category including nouns, pronouns and other word-classes. It is argued here that difficulties relating to the boundaries of Hudson’s ‘noun’ category demonstrate that there is no uncontroversial way to derive such a superordinate category from POS tagging. Decisions regarding the boundary of the ‘noun’ category have small but statistically significant effects on the ratio that emerges for ‘nouns’ as a whole. Tokenisation and categorisation differences between tagging schemes make it problematic to compare the ratio of ‘nouns’ across different tagsets. The precise figures for POS ratios are therefore effectively artefacts of the tagset. However, these objections to the use of POS ratios do not apply to their use as a metric of variation for comparing datasets tagged with the same tagging scheme.

Una grande varietà di input sensoriali dal mondo e dal corpo, sono continuamente integrati nel cervello al fine di creare rappresentazioni mentali sovramodali e coerenti del nostro stesso corpo. La plasticità è una caratteristica fondamentale di tali rappresentazioni, che consente costanti cambiamenti adattativi nelle funzioni mentali e nel comportamento. Anche le rappresentazioni corporee possono cambiare in base all'esperienza e, soprattutto, possono essere temporaneamente modificate mediante protocolli sperimentali. Nel presente lavoro, eravamo interessati a valutare la plasticità della percezione metrica del corpo e l'effetto di cambiamenti temporanei in essa sull'elaborazione delle informazioni corporee e spaziali. A tale scopo, sono stati utilizzati due illusioni corporee: la Mirror Box Illusion (MB) e la Full-Body Illusion (FBI). Il meccanismo principale che spiega l'efficacia di queste procedure sperimentali è il processo di incorporazione di una parte del corpo aliena. Nell'esperimento 1 abbiamo usato un paradigma visuo-tattile di FBI per valutarne la fattibilità e la replicabilità con corpi di dimensioni diverse. Abbiamo confermato che è possibile indurre e replicare nello stesso partecipante l'incorporazione verso manichini di dimensioni standard o più grandi. Nell'esperimento 2 e 3 abbiamo studiato la rappresentazione metrica della gamba e la sua malleabilità. Abbiamo quindi misurato l'effetto dell'FBI indotto da diverse dimensioni corporee, su un compito di valutazione della distanza percepita tra due tocchi applicati alla gamba del partecipante. Abbiamo scoperto che l'esperienza soggettiva di incorporazione è accompagnata da un cambiamento nella percezione della metrica del corpo che va di pari passo con la dimensione delle gambe incarnate. Poiché abbiamo confermato che, in soggetti sani, la rappresentazione metrica del corpo può essere modulata, abbiamo affrontato una domanda simile in pazienti con emiplegia. Nell'esperimento 4, usando un compito di bisezione del corpo abbiamo osservato che pazienti emiparetici mostrano una distorsione prossimale nella rappresentazione metrica dell'arto interessato. Abbiamo, inoltre, scoperto che la bisezione si sposta verso il punto medio reale dopo una sessione di trattamento con MB, rispetto a un trattamento di controllo senza specchio. Nell'esperimento 5 abbiamo trovato una modulazione simile della metrica corporea che, in un gruppo di pazienti affetti da aprassia ideomotoria trattati con una versione modificata della MB, era accompagnata da un miglioramento della programmazione dei piani motori. Negli esperimenti 6 e 7 ci siamo concentrati maggiormente sulla relazione tra metrica del corpo e rappresentazione dello spazio. In primo luogo, abbiamo testato l'ipotesi che una rappresentazione del corpo alterata influenzasse la percezione delle proprie attività motorie immaginate. I risultati hanno mostrato che i partecipanti immaginavano di camminare più velocemente dopo essere stati esposti a una FBI con gambe più lunghe. Inoltre, abbiamo scoperto che l'incorporazione illusoria di gambe più lunghe può influenzare la stima delle distanze allocentriche nello spazio extra-personale. L'incorporazione di gambe più lunghe, da un lato, ha, infatti, ridotto la distanza percepita in metri, dall'altro, ha prodotto un aumento del numero di passi che i partecipanti immaginavano di dover percorrere tra gli stessi punti di riferimento. In conclusione, abbiamo confermato che è possibile manipolare la rappresentazione metrica del corpo, mediante illusioni corporee e che ciò influenza la nostra capacità di stimare le distanze nel mondo esterno sia in termini di raggiungibilità che di stima allocentrica della distanza. Tale plasticità della rappresentazione corporea e dell'interazione spazio-corpo fornisce importanti indizi per la comprensione della rappresentazione corporea e della sua riabilitazione nei pazienti neurologici.
A large variety of sensory input from the world and the body, are continuously integrated in the brain in order to create supra-modal and coherent mental representations of our own body. Plasticity is a fundamental characteristic of the nervous systems, allowing constant adaptive changes in mental functions and behaviour. Thanks to this, even body representations can change according to experience and, crucially, they can be temporarily altered by means of experimental protocols. In the present work, we were interested in assessing the plasticity of the subjective metric of the body, and the effect of temporary changes in it on the processing of corporeal and spatial information. To this aim, two types of bodily illusion were used, i.e. the Mirror Box Illusion (MB) and the Full-Body Illusion (FBI), due to their known effects inducing strong modulations of body representation. The core mechanism accounting for the efficacy of these experimental procedures is likely to be the process of embodiment of an alien body part. In experiment 1 we used a visuotactile FBI-like paradigm to assess the feasibility and the replicability of the FBI for bodies of different sizes. Using this paradigm, we confirmed that it is possible to induce and replicate in the same participant, the embodiment towards mannequins of standard or bigger sizes. In experiment 2 and 3 we investigated body metric representation of the leg, and whether it can be plastically modulated by embodying mannequins of different sizes. To address this issue, we measured the effect of FBI induced by different body sizes, over a Body Distance Task (BDT), i.e. the assessment of the perceived distance between two touches applied to the participant’s leg. We found that the subjective experience of embodiment is also accompanied by a change in the perception of body metric that goes hand-in-hand with the current size of the embodied legs. Since we confirmed that, in healthy subjects, the metric representation of the body can be modulated, we addressed a similar question in patients with hemiplegia. In experiment 4, using a body bisection task we first observed that hemiparetic post-stroke patients show a proximal bias in the metric representation of their affected upper limb. Critically, we found that this bias shifts distally, towards the objective midpoint after a MB training session, compared to a control training without the mirror. In Experiment 5 we found a similar modulation of subjective body metric in a group of patients suffering from Ideomotor Apraxia, treated with a modified version of the MB setup, which was accompanied by an improvement in the programming of motor plans. In experiments 6 and 7 we focused more on the relationship between body metric and space representations. First, we tested the hypothesis that an altered body representation could modify the way in which individuals estimate their body affordances during a Motor Imagery Task. Our results showed that participants imagined walking faster after having been exposed to an illusion of longer legs. Furthermore, we found that the illusory embodiment of longer legs can affect the estimation of allocentric distances in extra-personal space. The embodiment of longer legs, on the one hand, reduced the perceived distance in meters, on the other hand, produced an enhancement of the number of steps that participants imagined they would have needed to walk between the same landmarks. In conclusion, we confirmed that it is possible to induce provisional modifications of the metric representation of the body, by means of body illusions. We showed that body representation is malleable to the point to shape our ability to estimate distances in the external world both in terms of reachability and allocentric distance estimation. Such plasticity of body representation and body-space interaction gives important clues for the understanding of body representation and its rehabilitation in neurological patients.

La conoscenza della grandezza delle proprie parti corporee è essenziale per muoversi in maniera efficiente nell’ambiente esterno e per interagire accuratamente sia con gli oggetti sia con le altre persone. Attraverso un approccio interdisciplinare che combina paradigmi neurofisiologici (stimolazione cerebrale non invasiva) e comportamentali, la presente tesi indaga i meccanismi cognitivi e neurali sottostanti la rappresentazione della grandezza delle parti corporee. Lo Studio #1 dimostra il ruolo causale della corteccia somatosensoriale primaria nell’elaborazione della grandezza delle proprie parti del corpo. Nei soggetti adulti neurologicamente sani, la Stimolazione Magnetica Transcranica ripetitiva a 1-Hz della rappresentazione della mano nella mappa somatosensoriale di entrambi gli emisferi, induce delle distorsioni percettive (sovrastima) della grandezza della propria mano – come valutato con un compito visuo-percettivo – che non si estendono ad altri distretti corporei (il piede). Invece, cambiamenti nell’eccitabilità corticale indotti da Stimolazione Magnetica Transcranica ripetitiva del lobulo parietale inferiore destro o sinistro non influenzano la stima percettiva della grandezza della propria mano. Tale evidenza sottolinea il coinvolgimento causale della corteccia somatosensoriale primaria nella costruzione e nell’aggiornamento della rappresentazione metrica del proprio corpo. Lo Studio #2 si focalizza sui cambiamenti plastici che avvengono manipolando il senso di appartenenza corporea mostrando che, negli adulti neurologicamente sani, l’embodiment di mani più grandi (ma non più piccole) della propria influenza la rappresentazione percettiva cosciente della dimensione della propria mano. Infine, comparando la rappresentazione metrica del corpo in bambini a sviluppo tipico con quella degli adulti neurologicamente sani, lo Studio #3 mostra come le distorsioni percettive della rappresentazione corporea emergono durante il corso dello sviluppo. Complessivamente, i risultati della presente tesi supportano la natura estremamente flessibile della rappresentazione metrica del proprio corpo, mostrando come le distorsioni plastiche della grandezza delle proprie parti corporee si sviluppano gradualmente nell’arco della vita e possono essere modulate sia da cambiamenti neurofisiologici, sia da manipolazioni illusorie del senso di embodiment.
The knowledge of the size of the own body-parts is essential for efficiently moving in the external environment and accurately interacting both with objects and with other people. In an interdisciplinary approach which combines neurophysiological (i.e., non-invasive brain stimulation) and behavioral paradigms, the present dissertation investigates the cognitive and neural signatures underlying the representation of body-parts size. Study #1 demonstrates the casual role of the primary somatosensory cortex in one’s own body-parts size processing. In healthy adults, 1-Hz repetitive Transcranial Magnetic Stimulation over the hand representation in the somatosensory map of both hemispheres leads to perceptual distortions (i.e., overestimation) of the own hand size – as assessed with a visual perceptual task – which do not extend to other body districts (namely, the foot). Instead, cortical excitability shifts induced by repetitive Transcranial Magnetic Stimulation over the right or left inferior parietal lobule do not affect the perceptual estimation of the own hand size. This evidence highlights the causal involvement of the primary somatosensory cortex in the construction and updating of one’s own body metric representation. Study #2 focuses on the plastic changes which occur by manipulating the sense of body ownership, showing that, in healthy adults, the embodiment of external hands bigger (but not smaller) than the own affects the perceptual conscious representation of the own hand dimension. Finally, by comparing body metric representation in typically developing children and healthy adults, Study #3 shows how perceptual distortions of body-parts representation arise during the developmental course. Overall, findings from this dissertation support the extremely flexible nature of one’s own body metric representation, showing how plastic distortions of the own body-parts size develop gradually during the lifespan and can be modulated by neurophysiological changes as well as by illusory manipulations of self-attribution.

O tamanho corporal é uma importante característica, de indivíduos até comunidades e diz respeito ao crescimento, reprodução, mortalidade e alimentação. Pode ser utilizado para avaliar alterações no funcionamento e estrutura dos ecossistemas. Este trabalho é dividido em dois capítulos. No primeiro, fizemos uma revisão bibliográfica para investigar quais fatores de alteração ambiental afetam o tamanho corporal de peixes. Utilizamos uma metodologia semelhante a uma revisão sistemática. Identificamos oito fatores antrópicos de alteração ambiental que influenciam o tamanho corporal de peixes, no entanto os resultados da revisão não apontaram um padrão de resposta do tamanho corporal e poucos artigos selecionados na revisão utilizaram mais de uma métrica para medir o tamanho e este fato pode ser fundamental para um melhor entendimento dos efeitos decorrentes das alterações ambientais. No segundo, avaliamos como a variação do tamanho corporal em grupos tróficos em um sítio responde às alterações resultantes da agricultura quantificadas em quatro escalas (local, riparia até 1km a montante, riparia toda montante e em subbacia). No segundo capítulo investigamos como o tamanho corporal entre grupos tróficos de peixes responde as alterações decorrentes da agricultura em 4 escalas de paisagem em 52 riachos do Pampa Os resultados sugerem que a variação no tamanho corporal depende da medida de distribuição de tamanhos utilizada, do grupo trófico e da escala onde a perturbação é avaliada. A maioria dos grupos tróficos apresentou uma relação negativa com a escala local, geralmente com a redução do tamanho corporal médio ou com uma maior representação de tamanhos pequenos (left skewed). Enquanto em escalas riparia ou de bacia, tamanhos intermediários prevaleceram e tamanhos grandes e pequenos foram desfavorecidos (valores de curtose positivos). Nós ressaltamos a importância de se considerar as características tróficas das espécies para avaliações de alterações ambientais, uma vez que os artigos selecionados na revisão poucas vezes discutiam questões a respeito do efeito das alterações na estrutura trófica das assembléias de peixes. Além disso, a avaliação das escalas de agricultura reforça a importância da conservação da vegetação riparia. O uso de diferentes medidas de distribuição é importante para entender os efeitos da agricultura no tamanho corporal, uma vez que curtose, skewness e coeficiente de variação podem variar sem haver uma mudança no tamanho corporal médio dos peixes.
Body size is an important feature from individuals to communities and it concerns to growth, reproducing, mortality and diet. Body size can be used to evaluate variation on the ecosystem structure. This study was divided in two chapters. In the first, we performed a literature review to investigate which factors of environmental variations affect body size of fishes. We identified eight anthropic factors of environmental disturbance that influence fishes body size, however the results did not show any pattern related to body size variation and few selected papers used more than one size metric what could be fundamental to better understanding the effects caused by environmental changes. In the second chapter, we evaluated how the body size variation in trophic groups responds to agriculture in four landscape scales in 52 streams of the Pampas. The results suggested that body size variation depends on distributional size measures, on trophic group and on scale. Most of the trophic groups have negative relationship with to local scale, usually with mean body size reduction or with higher concentration of small sizes (left skewed), while in the riparian and basin scales fishes with mean values prevailed, and fishes with bigger and smaller sizes had disadvantage (positive values of kurtosis). We highlight the importance of considering trophic features of species to evaluate environmental changes, once previously selected papers rarely discussed the effect of changes on trophic structure of fish communities. Besides, evaluation of agriculture scales reinforce the importance of conservation of riparian vegetation. The use of different distributional measures is important to understand the effects of agriculture on body size, once kurtosis, skewness and coefficient of variation could variate without having changes on mean body size of fishes.

Although tracking data have completely revolutionized the whole data science paradigm in sports competitions with the largest economic resources, its use in a European context is still unexplored. In this thesis, three tracking-related contributions are presented in the sports domain. First, the creation of vision-based basketball multi-tracking methods is studied from a single-camera perspective, which could be useful for clubs with low resources or for the recovery of vintage games’ tracking. Then, tracking data in the soccer domain is enriched by adding a novel layer of information: player body-orientation, thus complementing 2D location data, which falls short in some scenarios. Finally, the effect of proper orientation is detailed in the most common soccer action: passes. By building passing computational models that express which is the safest pass at a given time, the relevance of orientation is contextualized, hence proving that it is indeed a vital skill for soccer players.
Tot i que les dades de seguiment han revolucionat el paradigma de la ciència de dades esportiva dins les competicions amb més recursos, el seu ús en un context europeu és encara una incògnita. En aquesta tesi, presentem tres contribucions dins d’aquest camp. Primer s’ha estudiat, a través de la visió per computador, la creació de sistemes de seguiment de jugadors/es de bàsquet utilitzant una sola càmera, el que podria servir per equips amb pocs recursos o per recuperar dades de partits antics. A més, donat que la manca de context és la principal limitació de les dades posicionals, la segona proposta en presenta l’enriquiment amb una nova capa d’informació: l’orientació corporal de jugadors/ es de futbol. Finalment, s’ha analitzat l’impacte de l’orientació mitjançant la creació de models computacionals de passades, els quals esbrinen quina és la passada més viable i demostren que l’orientació és una capacitat clau per als jugadors/es.

Submitted by PPG Ci?ncia da Computa??o (ppgcc@pucrs.br) on 2018-07-26T19:48:07Z No. of bitstreams: 1 CARLOS ALBERTO DOS SANTOS_DIS.pdf: 1271475 bytes, checksum: 856ae87ad633d3c772b413816caa43d1 (MD5)
Approved for entry into archive by Sheila Dias (sheila.dias@pucrs.br) on 2018-08-01T13:39:36Z (GMT) No. of bitstreams: 1 CARLOS ALBERTO DOS SANTOS_DIS.pdf: 1271475 bytes, checksum: 856ae87ad633d3c772b413816caa43d1 (MD5)
Made available in DSpace on 2018-08-01T14:31:21Z (GMT). No. of bitstreams: 1 CARLOS ALBERTO DOS SANTOS_DIS.pdf: 1271475 bytes, checksum: 856ae87ad633d3c772b413816caa43d1 (MD5) Previous issue date: 2018-04-27
It is known that linguistic processing of corpora demands high computational effort because of the complexity of its algorithms, but despite this, the results reached are better than that generated by the statistical processing, where the computational demand is lower. This dissertation describes a comparative analysis between the process linguistic and statistical of term extraction. Experiments were carried out through four corpora in English idiom, built from scientific papers, on which terms extractions were carried out using the approaches. The resulting terms lists were refined with use of relevance metrics and stop list, and then compared with the reference lists of the corpora across the recall technical. These lists, in its turn, were built from the context these corpora, whith help of Internet searches. The results shown that the statistical extraction combined with the stop list and relevance metrics can produce superior results to linguistic process extraction using the same metrics. It?s concluded that statistical approach composed by these metrics can be ideal option to relevance terms extraction, by requiring few computational resources and by to show superior results that found in the linguistic processing.
Sabe-se que o processamento lingu?stico de corpora demanda grande esfor?o computacional devido ? complexidade dos seus algoritmos, mas que, apesar disso, os resultados alcan?ados s?o melhores que aqueles gerados pelo processamento estat?stico, onde a demanda computacional ? menor. Esta disserta??o descreve uma an?lise comparativa entre os processos lingu?stico e estat?stico de extra??o de termos. Foram realizados experimentos atrav?s de quatro corpora em l?ngua inglesa, constru?dos a partir de artigos cient?ficos, sobre os quais foram executadas extra??es de termos utilizando essas abordagens. As listas de termos resultantes foram refinadas com o uso de m?tricas de relev?ncia e stop list, e em seguida comparadas com as listas de refer?ncia dos corpora atrav?s da t?cnica do recall. Essas listas, por sua vez, foram constru?das a partir do contexto desses corpora e com ajuda de pesquisas na Internet. Os resultados mostraram que a extra??o estat?stica combinada com as t?cnicas da stop list e as m?tricas de relev?ncia pode produzir resultados superiores ao processo de extra??o lingu?stico refinado pelas mesmas m?tricas. Concluiu se que a abordagem estat?stica composta por essas t?cnicas pode ser a op??o ideal para extra??o de termos relevantes, por exigir poucos recursos computacionais e por apresentar resultados superiores ?queles encontrados no processamento lingu?stico.

The purpose of this dissertation is to propose a reliability metric and respective validation tools for corpora annotated with deep linguistic information. The annotation of corpus with deep linguistic information is a complex task, and therefore is aided by a computational grammar. This grammar generates all the possible grammatical representations for sentences. The human annotators select the most correct analysis for each sentence, or reject it if no suitable representation is achieved. This task is repeated by two human annotators under a double-blind annotation scheme and the resulting annotations are adjudicated by a third annotator. This process should result in reliable datasets since the main purpose of this dataset is to be the training and validation data for other natural language processing tools. Therefore it is necessary to have a metric that assures such reliability and quality. In most cases, the metrics uses for shallow annotation or parser evaluation have been used for this same task. However the increased complexity demands a better granularity in order to properly measure the reliability of the dataset. With that in mind, I suggest the usage of a metric based on the Cohen’s Kappa metric that instead of considering the assignment of tags to parts of the sentence, considers the decision at the level of the semantic discriminants, the most granular unit available for this task. By comparing each annotator’s options it is possible to evaluate with a high degree of granularity how close their analysis were for any given sentence. An application was developed that allowed the application of this model to the data resulting from the annotation process which was aided by the LOGON framework. The output of this application not only has the metric for the annotated dataset, but some information related with divergent decision with the intent of aiding the adjudication process.

Tese de mestrado em Engenharia Informática, apresentada à Universidade de Lisboa, através da Faculdade de Ciências, 2011
Grande parte das ferramentas de processamento de linguagem natural utilizadas hoje em dia, desde os anotadores morfossintácticos (POS taggers) até aos analisadores sintáticos (parsers), necessita de corpora anotados com a informação linguística necessária para efeitos de treino e avaliação. A qualidade dos resultados obtidos por estas ferramentas está directamente ligada à qualidade dos corpora utilizados no seu treino ou avaliação. Como tal, é do mais alto interesse construir corpora anotados para treino ou avaliação com o maior nível de qualidade. Tal como as técnicas e as ferramentas da área do processamento de linguagem natural se vão tornando mais sofisticadas e tecnicamente mais complexas, também a quantidade e profundidade da informação contida nos corpora anotados tem vindo a crescer. O estado da arte actual consiste em corpora anotados com informação gramatical profunda, isto é anotação que contém não só a função ou tipo de cada elemento mas também os tipos das relações entre os diferentes elementos, sejam estas directas ou de longa distância. Esta quantidade crescente de informação contida na anotação dos corpora torna a tarefa da sua anotação crescentemente mais complexa, daí existir a necessidade de garantir que este processo resulta em corpora da melhor qualidade possível. No seguimento desta crescente complexidade, as técnicas utilizadas para o processo de anotação também tem sofrido alterações. A quantidade de informação a ser introduzida no corpus é demasiado complexa para ser introduzida manualmente, portanto este processo é agora conduzido por uma gramática computacional, que produz todas as possíveis representações gramaticais para cada frase, e de seguida um ou mais anotadores humanos escolhem a representação gramatical que melhor se aplica a frase em questão. Este processo garante uma uniformidade no formato da anotação, bem como consistência total nas etiquetas utilizadas, problemas recorrentes em corpus anotados manualmente. O objectivo desta dissertação é o de identificar um método ou uma métrica que possibilite a avaliação da tarefa de anotação de corpora com informação gramatical profunda, bem como uma aplicação que permita a recolha dos dados necessários referentes à tarefa de anotação, e que calcule a métrica ou métricas necessárias para validação e avaliação da tarefa. Com este objectivo em mente, foi inicialmente explorado o trabalho de fundo da tarefa de anotação, tanto na vertente linguística como na vertente de processamento de linguagem natural. Na vertente linguística, devem ser realçadas algumas noções base, tais como a de corpus, que se trata de um acervo de material linguístico originário de múltiplas fontes, tais como emissões de rádio, imprensa escrita e até conversas do dia-a-dia. Um corpus anotado é um corpus em que o material foi explicitamente enriquecido com informação linguística que é implícita para um falante nativo da língua, com o objectivo de auxiliar ao processamento do material por parte de máquinas. A anotação de corpus por parte do grupo NLX está a ser feita recorrendo a um esquema de anotação duplamente cego, em que dois anotadores escolhem de um conjunto de possíveis representações gramaticais atribuídas a cada frase pela gramática LXGram, a que para si é a mais correcta. Estas representações são posteriormente adjudicadas por um terceiro anotador. O resultado desta adjudicação é a representação que integra o corpus anotado. O foco deste trabalho é o de avaliar a qualidade e fiabilidade do material resultante deste processo de anotação. O processo de anotação pode ser visto como o processo de atribuição de categorias a itens, neste caso, a atribuição de categorias ou informação linguística a palavras ou multi-palavras de uma frase. Neste caso concreto, dada uma lista de discriminantes semânticos, os anotadores devem decidir quais pertencem ou não à melhor representação gramatical de uma dada frase. Na literatura, existem várias abordagens para a avaliação de anotação com esquemas de anotação simples, por exemplo, com anotação morfossintáctica (POS tagging), como é o caso do Cohen’s Kappa (Cohen, 1960), ou k, e suas variantes, tais como o S (Bennett et al., 1954), _ (Scott, 1955) ou o próprio k. Todas estas métricas se baseiam na mesma ideia de que a taxa de concordância entre anotadores (inter-annotator agreement) pode ser calculada tendo em conta dois valores: a concordância observada (Ae), isto é a quantidade de informação em relação à qual os anotadores concordam; e a concordância esperada (Ao), ou seja a quantidade de informação que se esperaria obter entre os anotadores se a anotação fosse feita aleatoriamente. Todas as métricas derivadas directamente do Cohen’s Kappa, calculam também a taxa de concordância da mesma forma, recorrendo à fórmula: concordância = Ao–Ae 1–Ae. O ponto de divergência entre as diferentes abordagens está na maneira de calcular a taxa de concordância esperada. Estas divergências consistem na representação da taxa de concordância esperada através de diferentes distribuições estatísticas. Existe outro tipo de métricas, normalmente utilizado para a avaliação de análises sintáticas que também são aplicadas neste tipo de tarefa. Métricas como são o caso do Parseval (Black et al., 1991) e do Leaf Ancestor (Sampson and Babarczy, 2003) que frase a frase comparam a análise sintática dada pelo analisador sintático automático com um padrão dourado (análise sintática considerada correcta para a frase). Contudo, a complexidade da tarefa a ser avaliada exige não só uma métrica sólida, mas também que a sua granularidade seja suficiente para distinguir pequenas divergências que podem sustentar resultados que aparentam ser contraditórios. Tendo em conta a tarefa a ser avaliada, a abordagem mais granular possível é a que consiste em comparar individualmente cada decisão sobre cada discriminante para uma dada frase. Portanto, visto que o objectivo é obter a maior granularidade possível, para a métrica desenvolvida Y-Option Kappa, a taxa de acordo observado pode ser calculada pela razão entre o número de discriminantes com decisões idênticas, ou opções, e o número total de discriminantes disponíveis para uma dada frase. Como cada discriminantes tem dois valores possíveis, isto é, ou pertence ou não à melhor representação gramatical, a taxa de concordância esperada pode ser considerada uma distribuição uniforme de decisões binárias, o que significa que o acordo esperado para caso de decisão aleatória será 0,5. A métrica Y-Option Kappa é calculada através da mesma fórmula utilizada pelo Cohen’s K e suas variantes. A tarefa de anotação é auxiliada por um pacote de ferramentas linguísticas designado LOGON, pacote este que permite a anotação dinâmica de corpus, isto é as frases são analisadas dinamicamente pela gramática computacional conforme as decisões sobre os discriminantes são tomadas pelos anotadores. Isto permite ter acesso às representações gramaticais resultantes, possibilitando assim uma melhor percepção do resultado das decisões tomadas. A informação resultante do processo de anotação é guardada em ficheiros de log que podem ser utilizados para reconstruir a representação gramatical resultante para a frase. Este pacote é bastante útil e fornece uma ajuda preciosa no processo de anotação. Contudo, os ficheiros de log guardam apenas a informação necessária para a reconstrução da representação gramatical final, o que resulta numa lista de discriminantes que pode ser incompleta para os propósitos de avaliação do processo de anotação. Por exemplo, quando um anotador rejeita uma frase, ou seja, considera que não existe no conjunto possível de representações gramaticais uma que seja considerada correcta, apenas os discriminantes considerados até ao momento da rejeição são registados no ficheiro de log. Para resolver este problema, algumas adaptações tiveram de ser feitas à ideia original da métrica Y-Options K para que esta fosse aplicável aos dados recolhidos. Existem três casos gerais que resultam em conjuntos de informação concretos nos ficheiros de log. Estes três casos são: • Cada anotador aceita uma representação gramatical como óptima para a frase: Todas as opções estão presentes e podem ser comparadas correctamente • Pelo menos um dos anotadores rejeita qualquer representação gramatical para a frase: Existe apenas uma lista parcial das opções tomadas (para esse anotador). Para resolver estes casos, são estimados sobre os casos em que toda a informação está disponível valores médios que são depois aplicados a casos em que a informação não esteja disponível. A métrica é assim calculada frase a frase, e o resultado final apresentado é a média aritmética da métrica para todas as frases. Foi desenvolvida uma aplicação que permite através dos ficheiros de log determinar o valor da métrica, bem como alguma informação adicional para auxílio da tarefa de adjudicação. Um objectivo futuro seria o de alterar as aplicações do pacote LOGON, mais concretamente o [incr tsdb()] de modo a que este guarde todos os discriminantes para cada frase, podendo assim dispensar o cálculo de estimativas.
The purpose of this dissertation is to propose a reliability metric and respective validation tools for corpora annotated with deep linguistic information. The annotation of corpus with deep linguistic information is a complex task, and therefore is aided by a computational grammar. This grammar generates all the possible grammatical representations for sentences. The human annotators select the most correct analysis for each sentence, or reject it if no suitable representation is achieved. This task is repeated by two human annotators under a double-blind annotation scheme and the resulting annotations are adjudicated by a third annotator. This process should result in reliable datasets since the main purpose of this dataset is to be the training and validation data for other natural language processing tools. Therefore it is necessary to have a metric that assures such reliability and quality. In most cases, the metrics uses for shallow annotation or parser evaluation have been used for this same task. However the increased complexity demands a better granularity in order to properly measure the reliability of the dataset. With that in mind, I suggest the usage of a metric based on the Cohen’s Kappa metric that instead of considering the assignment of tags to parts of the sentence, considers the decision at the level of the semantic discriminants, the most granular unit available for this task. By comparing each annotator’s options it is possible to evaluate with a high degree of granularity how close their analysis were for any given sentence. An application was developed that allowed the application of this model to the data resulting from the annotation process which was aided by the LOGON framework. The output of this application not only has the metric for the annotated dataset, but some information related with divergent decision with the intent of aiding the adjudication process.

This chapter describes an academic and rigorous evaluation of the utility and current short-comings of state-of-the-art game AI to support junior leadership training outcomes in the defence domain. The chapter describes the design and implementation of a number of section-level (9 soldiers, one of which is the junior leader – typically a corporal) scenarios in the serious-game/military-simulation known as VBS (Virtual Battlespace 2). A number of objective experiments are conducted to quantify the utility of AI for junior leadership training. A suite of performance metrics were implemented using VBS2’s scripting capabilities. These metrics included such scorings as loss-exchange-ratios, number of rounds expended, time to complete mission, distribution (by role) of casualties within the section, et cetera.

This study investigates the production of and exposure to lexical features when non-native speakers (NNS) converse with each other (NNS-NNS) engaging in interlanguage talk, as compared to when they engage in naturalistic speech with a native speaker (NS). The authors focus on lexical features that are associated with breadth of lexical knowledge including lexical diversity and lexical frequency. Spoken corpora from three types of dyads (NS-NNS, NNS-NS, NNS-NNS) are analyzed using the computational tool, Coh-Metrix. The results indicate that NNSs produce language with significantly greater lexical diversity and higher word frequency (i.e., more common words) when speaking to another NNS than when speaking to a NS. Hence, there is greater breadth of lexical knowledge apparent within interlanguage dyads (i.e., NNS-NNS) than within NNS-NS dyads in the variety of words produced, but not the frequency of the words. There were no significant differences in NNS exposure to breadth of lexical knowledge features as a function of whether the speaker was a NS or NNS. Hence, NNSs were exposed to similar levels of lexically comprehensible input regardless of interlocutor. These findings have important implications for the developmental role of interlanguage talk in reference to lexical production and exposure.

Text detoxification is the task of rewriting a toxic text into a neutral text while preserving its original content. It has a wide range of applications, e.g. moderation of output of neural chatbots or suggesting less emotional version of posts on social networks. This paper provides a description of RUSSE-2022 competition of detoxification methods for the Russian language. This is the first competition which features (i) parallel training data and (ii) manual evaluation. We describe the setup of the competition, the solutions of the participating teams and analyse their performance. In addition to that, the large-scale evaluation allows us to analyse the performance of automatic evaluation metrics.

This paper investigates the problem of creating summaries of Russian-language texts based on extractive (TextRank and LexRank) and abstractive (mBART, ruGPT3Small, ruGPT3Large, ruT5-base and ruT5-large) methods. For our experiments, we used the Russian-language corpus of news articles Gazeta and the Russianlanguage parts of the MLSUM and XL-Sum corpora. We computed ROUGE-N, ROUGE-L, BLEU, METEOR and BERTScore metrics to evaluate the quality of summarization. According to the experimental results, the methods are ranked (from best to worst) as follows: ruT5-large, mBART, ruT5-base, LexRank, ruGPT3Large, TextRank, ruGPT3Small. The study also highlights the salient features of summaries obtained by various methods. In particular, mBART summaries are less abstractive than ruGPT3Large and ruT5-large, and ruGPT3Large summaries are often incomplete and contain errors.

A novel real-time algorithm has been developed for estimating temporal word boundaries in measured speech without the need for interpretation of individual words. This algorithm is the foundational building block of a method for estimating a variety of key metrics such as word production rate, phrase production rate, words per phrase, etc., that are indicative of human mental states. In particular, we are interested in developing a system for monitoring locomotive crew alertness. The majority of existing speech processing algorithms relies on pre-recorded speech corpora. The real-time algorithm presented here is unique in that it employs a simple and efficient pattern matching method to identify temporal word boundaries by monitoring threshold crossings in the speech power signal. This algorithm eliminates the need to interpret the speech, and still produces reasonable estimates of word boundaries. The proposed algorithm has been tested with a batch of experimentally recorded speech data and with real time speech data. The results from the testing are outlined in this paper.

Abstract: Every e-learning teacher proposes presentation slides to learners who often stem from larger and more complex lecture notes. Delivering both these learning formats has become the cornerstone of every university e-learning course, and even MOOCs (Massive Open Online Courses) are often based on them. However, discrepancies between these formats in terms of complexity have not been quantitatively analyzed so far. This study aims at performing a detailed comparison between these modalities using variations of textual complexity metrics as background, ranging from surface, syntactic, morphological and semantic factors. The analyzed corpora are automatically extracted from MOOC materials and encompass multiple domains: history, politics, geography and culture, in order to induce diversity and to observe different inter-domain presentation characteristics. As an overview, there are high variations in terms of proportions (slides per lecture note page), underlying structures and individual word complexities. Although words on slides generally tend to be less complex than on lecture notes (nor too simple), the significant reduce of stop words and of connectors shifts the balance in terms of frequently used quantitative complexity factors. If the initial decrease in individual word complexities should denote a decrease in the perceived difficulty, the latter computational perspective determined by the elimination of irrelevant words that usually tend to be small in length, artificially increases the perceived complexity level. Moreover, the lower cohesion between bullet items from presentations, in contrast to adjacent sentences from the same paragraph extracted from lecture notes, negatively influences the overall variation of the complexity scores. In the end, we conclude by providing a set of required metrics for supporting teachers in the adaptation of their learning materials.