This dissertation demonstrates that the optimal account of English phonology denies phonemic status to oral glides. That is, it shows that all instances of phonetic [y] and [w] are predictable by rule. These occurrences include the following: formative initial glides, such as those in yet and wet; post-consonant, pre-vocalic [w] in such forms as quit, guava, and white and post-consonant, pre-vocalic [y] in such forms as cute, few, million, onion, and champion; the [y] following the tense vowels in bite, beet, bate, and boy and the [w] following the tense vowels in bout, boot, boat, cute, and few; and, finally, the post-vocalic centering glide [h] in spa, cloth, beer [bihr], and bear. The new proposals, described and justified in Chapter III, have the effect of eliminating the glides [y] and [w] from the inventory of underlying phonemes of English. From this flows what is perhaps more significant: they render the feature [Syllabic] completely redundant in the lexical representations of English formatives.

Previous research in evolutionary linguistics has made wide use of artificial language learning (ALL) paradigms, where learners are taught artificial languages in laboratory experiments and are subsequently tested in some way about the language they have learnt. The ALL framework has proved particularly useful in the study of the evolution of language, allowing the manipulation of specific linguistic phenomena that cannot be isolated for study in natural languages. Furthermore, this framework can test the output of individual participants, to uncover the cognitive biases of individual learners, but can also be implemented in a cultural evolutionary framework, investigating how participants acquire and change artificial languages in populations where they learn from and interact with each other. In this thesis, I present a novel methodology for studying the evolution of language in experimental populations. In the artificial sign language learning (ASLL) methodology I develop throughout this thesis, participants learn manual signalling systems that are used to interact with other participants. The ASLL methodology combines features of previous ALL methods as well as silent gesture, where hearing participants must communicate using only gesture and no speech. However, ASLL provides several advantages over previous methods. Firstly, reliance on the manual modality reduces the interference of participants’ native languages, exploiting a modality with linguistic potential that is not normally used linguistically by hearing language users. Secondly, research in the manual modality offers comparability with the only current evidence of language emergence and evolution in natural languages: emerging sign languages that have evolved over the last century. Although the silent gesture paradigm also makes use of the manual modality, it has thus far seen little implementation into a cultural evolutionary framework that allows closer modelling of natural languages that are subject to the processes of transmission to new learners and interaction between language users. The implementation and development of ASLL in the present work provides an experimental window onto the cultural evolution of language in the manual modality. I detail a set of experiments that manipulate both linguistic features (investigating category structure and verb constructions) and cultural context, to understand precisely how the processes of interaction and transmission shape language structure. The findings from these experiments offer a more precise understanding of the roles that different cultural mechanisms play in the evolution of language, and further builds a bridge between data collected from natural languages in the early stages of their evolution and the more constrained environments of experimental linguistic research.

Title: On the Linguistic Structure of Emotional Meaning in Czech Author: Mgr. Kateřina Veselovská Department: Institute of Formal and Applied Linguistics Supervisor: Prof. PhDr. Eva Hajičová, DrSc., Institute of Formal and Applied Linguistics Keywords: emotional meaning, linguistic structure, sentiment analysis, opinion mining, evaluative language Abstract: This thesis has two main goals. First, we provide an analysis of language means which together form an emotional meaning of written utterances in Czech. Sec- ond, we employ the findings concerning emotional language in computational applications. We provide a systematic overview of lexical, morphosyntactic, semantic and pragmatic aspects of emotional meaning in Czech utterances. Also, we propose two formal representations of emotional structures within the framework of the Prague Dependency Treebank and Construction Grammar. Regarding the computational applications, we focus on sentiment analysis, i.e. automatic extraction of emotions from text. We describe a creation of manually annotated emotional data resources in Czech and perform two main sentiment analysis tasks, polarity classification and opinion target identification on Czech data. In both of these tasks, we reach the state-of-the-art results.

[pt] Este estudo investiga as habilidades linguísticas de crianças no processamento de estruturas linguísticas de alto custo. Foca-se nos possíveis efeitos do fechamento das escolas em decorrência da pandemia COVID-19 no desempenho de tarefas que envolvam estruturas de alto custo de processamento. Compara o desempenho linguístico de crianças de uma escola da rede pública de ensino do Rio de Janeiro em dois momentos: imediatamente após a reabertura das escolas, em 2022, após seu fechamento em decorrência da pandemia COVID-19 e um ano após a flexibilização do isolamento social, em 2023. Foram testados 100 escolares, com idade média de 9,5 anos e 8,8 anos, sem diagnóstico anterior de problemas de linguagem, por meio do módulo sintático da bateria MABILIN (Módulos de Avaliação de Habilidades Linguísticas), com vistas ao rastreio de dificuldades de linguagem, como as sugestivas do TDL (Transtorno do Desenvolvimento da Linguagem). No ano de 2022, das crianças testadas, 35,47 por cento apresentaram algum grau de dificuldade na compreensão das estruturas do MABILIN 1. Em contrapartida, em 2023, do total de crianças que apresentaram dificuldades em 2022, 61,1 por cento (22) não apresentaram mais dificuldades em 2023. Portanto, conclui-se ser possível que a pandemia COVID-19 tenha promovido dificuldades no desenvolvimento de habilidades linguísticas em escolares, embora não haja evidências claras de aumento da incidência de TDL. / [en] This study investigates schoolchildren s linguistic abilities in processing high-cost linguistic structures. It focuses on the possible effects of school closure due to the COVID-19 pandemic on the performance of tasks involving high-cost processing structures. The linguistic performance of children from a public school in Rio de Janeiro was compared at two different moments: immediately after the reopening of schools in 2022, after their closure due to the COVID-19 pandemic, and one year after the easing of social isolation, in 2023. One hundred students, with an average age of 9.5 years and 8.8 years, without prior diagnosis of language problems, were tested using the syntactic module of the MABILIN battery (Language Skills Assessment Modules), with a view to screening for language difficulties, such as those suggestive of DLD (Developmental Language Disorder). In 2022, 35.47 percent of the children tested presented some degree of difficulty in understanding the structures of MABILIN 1. This percentage differs considerably from what was obtained before the COVID-19 pandemic through the same battery of tests (7-9 percent), which is similar to what has been observed in different languages. In contrast, in 2023, of the total number of children who had difficulties in 2022, 61.1 percent (22) no longer had difficulties in 2023. Therefore, it is concluded that the COVID-19 pandemic may have promoted difficulties in the development of linguistic skills in schoolchildren, although there is no clear evidence of an increase in the incidence of DLD.

Contemporary approaches to natural language processing are predominantly based on statistical machine learning from large amounts of text, which has been manually annotated with the linguistic structure of interest. However, such complete supervision is currently only available for the world's major languages, in a limited number of domains and for a limited range of tasks. As an alternative, this dissertation considers methods for linguistic structure prediction that can make use of incomplete and cross-lingual supervision, with the prospect of making linguistic processing tools more widely available at a lower cost. An overarching theme of this work is the use of structured discriminative latent variable models for learning with indirect and ambiguous supervision; as instantiated, these models admit rich model features while retaining efficient learning and inference properties. The first contribution to this end is a latent-variable model for fine-grained sentiment analysis with coarse-grained indirect supervision. The second is a model for cross-lingual word-cluster induction and the application thereof to cross-lingual model transfer. The third is a method for adapting multi-source discriminative cross-lingual transfer models to target languages, by means of typologically informed selective parameter sharing. The fourth is an ambiguity-aware self- and ensemble-training algorithm, which is applied to target language adaptation and relexicalization of delexicalized cross-lingual transfer parsers. The fifth is a set of sequence-labeling models that combine constraints at the level of tokens and types, and an instantiation of these models for part-of-speech tagging with incomplete cross-lingual and crowdsourced supervision. In addition to these contributions, comprehensive overviews are provided of structured prediction with no or incomplete supervision, as well as of learning in the multilingual and cross-lingual settings. Through careful empirical evaluation, it is established that the proposed methods can be used to create substantially more accurate tools for linguistic processing, compared to both unsupervised methods and to recently proposed cross-lingual methods. The empirical support for this claim is particularly strong in the latter case; our models for syntactic dependency parsing and part-of-speech tagging achieve the hitherto best published results for a wide number of target languages, in the setting where no annotated training data is available in the target language.

Made available in DSpace on 2016-04-28T18:22:17Z (GMT). No. of bitstreams: 1 Fernanda de Castro Goncalves.pdf: 454064 bytes, checksum: 32a6dae975f10cad8e5b1fee4c330c63 (MD5) Previous issue date: 2011-05-26 / Secretaria da Educação do Estado de São Paulo / This research came into light due to the need to help my students produce written texts. The dissertational text is, in fact, an argumentative-dissertational text, for, according to Koch (1987), social interaction through language occurs mainly by argumentation. The difficulty in students´ communication, mainly the writing of the argumentative text, has always been present in teachers and researchers´discourses in the education field. Porta (2002), as a philosophy researcher is concerned with the relationship: Situation description- Problem- Thesis- Argumentation. Thus, a dissertational-argumentative text begins with a Description of a problem, proposes a Solution (actually, solution hypothesis) in favor of which the argumentation is presented. In case there are several hypotheses, he adds, it is the arguments that will select the best of them. As long as Description is concerned, he goes on saying that descriptions can play a leading role in many ways; what can´t be done is eliminating the problem as such, reducing the philosophical thesis to a mere description. Considering such questions, what I have observed in my student´s essays (I previously discuss the matter with them) is a tendency to describing the Situation (e.g. violence in São Paulo city), in that little attention is given to pointing out the Problem and because of that, there is no room for a Solution to the problem and being so, less attention is given to the Argumentation. Being a a systemicist, Eggins (1994) claims that the coherence of a text is based on two factors: (a) appropriateness to the genre, which involves a schematic structure (or genre structure) in its stages and finalities and (b) appropriateness to the linguistic register, which must comply with the variables of Field (subject matter), Tenor (the roles of the participants in an interaction) and Mode (how language organizes such elements). The objective of this research is to examine cohesion and coherence in dissertational- argumentative texts written by 2nd year high school students examining their linguistic structure and genre in order to check how they occur in their dissertational-argumentative texts to spot possible problems in these fields. My analysis is based upon Systemic Functional Linguistics ( developed by Halliday 1985; 1994 and his co-authors), which attempts to develop a theory about language as a social process and a methodological process that leads to a detailed and systemic description of the linguistic standards taking into account genre.Therefore, this research must answer the following research questions: (a) how is the genre structure configured in the examined essays? (b) How is the language in these essays presented? / dissertativoargumentativo, pois, como diz Koch (1987), a interação social por intermédio da língua caracteriza-se, fundamentalmente, pela argumentatividade. A dificuldade de comunicação escrita do aluno, em especial, nesse sentido, da redação do texto argumentativo, esteve sempre presente no discurso de professores e pesquisadores da área da educação. Porta (2002), de sua posição de estudioso da Filosofia, trata da relação: Descrição da Situação - Problema Tese Argumentação. Assim, um texto dissertativo-argumentativo, inicia-se com uma Descrição de um Problema, propõe-se uma Solução (na verdade, hipótese de Solução), em favor da qual apresenta-se a Argumentação. No caso de haver várias hipóteses, diz ele, são os argumentos que vão selecionar a melhor delas. A respeito da Descrição, diz ele que a descrição pode desempenhar um papel preponderante em vários sentidos; o que não pode é eliminar o problema enquanto tal, reduzindo, assim, uma tese filosófica a uma mera descrição. Com referências a essas questões, o que tenho verificado nas redações dos meus alunos (com quem sempre discuto previamente o assunto) é uma tendência à descrição da Situação (e.g. violência na cidade de São Paulo), em que pouca atenção de dá ao delineamento do Problema, e que, por isso, não dá lugar à apresentação de uma proposta (hipótese) de Solução, e, assim sendo, menos ainda para os Argumentos em defesa dessa proposta. Eggins (1994), de sua posição de sistemicista, afirma que a coerência de um texto repousa em dois fatores: (a) a adequação ao gênero, que envolve a estrutura esquemática (ou de gênero) com seus estágios e finalidades e (b) a adequação ao registro linguístico, que deve respeitar as variáveis de campo (assunto), relações (os interlocutores envolvidos) e modo (como a língua organiza esses elementos). O objetivo desta pesquisa é examinar a coerência e a coesão de textos dissertativo-argumentativos, escritos por alunos de 2º ano do Ensino Médio, examinando sua estrutura linguística e sua estrutura de gênero, a fim de verificar como elas se realizam em seus textos dissertativoargumentativos para detectar possíveis problemas nessas áreas. Assim, a pesquisa deve responder às seguintes perguntas de pesquisa: (a) Como se configura a estrutura de gênero nas redações examinadas? (b) Como se apresenta a linguagem dessas redações? Em minha análise, recorro à Linguística Sistêmico-Funcional, desenvolvida por Halliday (1985; 1994) e seus colaboradores, a qual procura desenvolver uma teoria sobre a língua como um processo social e uma metodologia que permita uma descrição detalhada e sistemática dos padrões linguísticos, tendo em vista o gênero

Contemporary approaches to natural language processing are predominantly based on statistical machine learning from large amounts of text, which has been manually annotated with the linguistic structure of interest. However, such complete supervision is currently only available for the world's major languages, in a limited number of domains and for a limited range of tasks. As an alternative, this dissertation considers methods for linguistic structure prediction that can make use of incomplete and cross-lingual supervision, with the prospect of making linguistic processing tools more widely available at a lower cost. An overarching theme of this work is the use of structured discriminative latent variable models for learning with indirect and ambiguous supervision; as instantiated, these models admit rich model features while retaining efficient learning and inference properties. The first contribution to this end is a latent-variable model for fine-grained sentiment analysis with coarse-grained indirect supervision. The second is a model for cross-lingual word-cluster induction and the application thereof to cross-lingual model transfer. The third is a method for adapting multi-source discriminative cross-lingual transfer models to target languages, by means of typologically informed selective parameter sharing. The fourth is an ambiguity-aware self- and ensemble-training algorithm, which is applied to target language adaptation and relexicalization of delexicalized cross-lingual transfer parsers. The fifth is a set of sequence-labeling models that combine constraints at the level of tokens and types, and an instantiation of these models for part-of-speech tagging with incomplete cross-lingual and crowdsourced supervision. In addition to these contributions, comprehensive overviews are provided of structured prediction with no or incomplete supervision, as well as of learning in the multilingual and cross-lingual settings. Through careful empirical evaluation, it is established that the proposed methods can be used to create substantially more accurate tools for linguistic processing, compared to both unsupervised methods and to recently proposed cross-lingual methods. The empirical support for this claim is particularly strong in the latter case; our models for syntactic dependency parsing and part-of-speech tagging achieve the hitherto best published results for a wide number of target languages, in the setting where no annotated training data is available in the target language.