This thesis presents novel tasks and deep learning methods for machine reading comprehension and question answering with the goal of achieving natural language understanding. First, we consider a semantic parsing task where the model understands sentences and translates them into a logical form or instructions. We present a novel semi-supervised sequential autoencoder that considers language as a discrete sequential latent variable and semantic parses as the observations. This model allows us to leverage synthetically generated unpaired logical forms, and thereby alleviate the lack of supervised training data. We show the semi-supervised model outperforms a supervised model when trained with the additional generated data. Second, reading comprehension requires integrating information and reasoning about events, entities, and their relations across a full document. Question answering is conventionally used to assess reading comprehension ability, in both artificial agents and children learning to read. We propose a new, challenging, supervised reading comprehension task. We gather a large-scale dataset of news stories from the CNN and Daily Mail websites with Cloze-style questions created from the highlights. This dataset allows for the first time training deep learning models for reading comprehension. We also introduce novel attention-based models for this task and present qualitative analysis of the attention mechanism. Finally, following the recent advances in reading comprehension in both models and task design, we further propose a new task for understanding complex narratives, NarrativeQA, consisting of full texts of books and movie scripts. We collect human written questions and answers based on high-level plot summaries. This task is designed to encourage development of models for language understanding; it is designed so that successfully answering their questions requires understanding the underlying narrative rather than relying on shallow pattern matching or salience. We show that although humans solve the tasks easily, standard reading comprehension models struggle on the tasks presented here.

The study investigated the relationship among three cognitive styles and summarization abilities. Both summarization products and processes were examined. Summarizing products were scored and a canonical correlation analysis was performed to determine their relationship with three cognitive styles. Summarizing processes were examined by videotaping students as they provided think aloud protocols. Their processes were recorded on composing style sheets and analyzed qualitatively. Subjects were sixth-grade students in self-contained classes in a suburban school district. Summarizing products were collected over a two week period in the fall. Summarizing processes were collected over an eight week period in the spring of the same school year. The results of the summarizing products analysis suggest that cognitive styles are related to summarization abilities. Two canonical correlations among the two variable sets were statistically significant at the .05 level of significance (.33 and .29). The results further suggest that students who are field independent, reflective, and flexible in their attentional style may be more adept at organizing their ideas and using written mechanics while summarizing. Students who are impulsive and constricted in attentional style may exhibit strength in expressing their ideas while summarizing. Results of the summarizing processes analysis suggest that students of one cognitive style combination may exhibit different behaviors while summarizing than those of other cognitive style combinations. Students who are field independent, reflective, and flexible in their attentional style seem to display more mature, interactive behaviors while summarizing than their peers of other cognitive style combinations.

Metacomprehension refers to the ability to judge one's own comprehension. Studies in the literature have shown that generating keywords after reading helps adults and children make comprehension judgments that are better correlated with their actual comprehension. Researchers have also found that when metacomprehension is framed in terms of confidence, there is an effect of ability, where individuals with low ability tend to be overconfident in their judgments, while those with high ability tend to be underconfident. This paper describes two experiments investigating metacomprehension in seventh graders. Experiment 1 sought to replicate and extend the finding that generating keywords after reading improves the accuracy of comprehension judgments and the effectiveness of study choices. To account for potential effects of time on task, participants in the control condition were asked to read passages twice in lieu of generating keywords. Two measures of metacomprehension accuracy (signed differences and gamma correlations) were based on comprehension judgments taken at two time points (pre-test and post-test). The moderating effects of reading ability were also examined. The results of Experiment 1 showed that participants were overconfident in their judgments of their own comprehension. Overconfidence was greater for pre-test predictions than for post-test reflections, and it was also greater for participants with lower reading ability. Generating keywords caused participants to become significantly less overconfident- or more accurate- from pre-test to post-test in their comprehension judgments, but it did not actually boost comprehension scores. In other words, generating keywords helped participants know that they did not know; it did not, however, help them know more. In Experiment 2, the investigation of generating keywords and rereading text was situated within a new context incorporating practice test questions. Studies have shown that practice testing is an effective study strategy. Additionally, since researchers have found that learners can use information about peer performance as a basis for making judgments about themselves, Experiment 2 also asked whether peer collaboration might increase metacomprehension accuracy. Participants were randomly assigned to four conditions: individual/keyword, individual/reread, collaborate/keyword, and collaborate/reread. All participants answered practice test questions; participants in the individual conditions worked on the questions alone, while participants in the collaborative conditions discussed the questions with a partner. As in Experiment 1, participants in Experiment 2 were also overconfident in judging their own comprehension. Again, there was an effect for time of judgment, such that predictions were more overconfident than were reflections. Surprisingly, peer collaboration was found to lead to greater overconfidence in comprehension judgments. Participants who collaborated with a peer were more overconfident than participants who worked alone. Experiment 2 showed that in the presence of practice testing and peer collaboration, the interactive effect of keyword generation and time of judgment was minimized. Within the keyword group, participants who collaborated and participants who worked alone did not differ in overconfidence. Within the reread group, however, participants who collaborated were significantly more overconfident than those who worked alone. Taken together, these two studies suggest that middle school students are generally overconfident in their judgments of comprehension. However, the results indicate that study strategies designed to enhance comprehension and learning can be effective in reducing students' overconfidence about themselves.

This dissertation examines how the reading of complex literary texts is enacted by select tenth-grade students, and their teachers, both within and outside of the classroom context, with an aim toward deconstructing "close reading" as a preferred pedagogical choice with insufficient theorization or supporting research. First, utilizing an individual think-aloud protocol, the researcher solicited the responses of nine students, and one of their tenth-grade English teachers, as they read for the first time three short story texts selected based on their identification by the Common Core State Standards as texts of complexity for high school students: Chekhov's Home, Poe's The Cask of Amontillado, and Borges' The Garden of Forking Paths. Those case study students were then studied ethnographically via the researcher's participant observation in their tenth-grade English classes for all days over the period of time that a major literary text was taught: Golding's Lord of the Flies. Based on the principles of microethnographic discourse analysis, the research applied open coding to all artifacts: the think-aloud commentaries, the verbatim transcripts of the audiotaped classroom oral discourse, the documents of classroom written discourse, and the verbatim transcripts of ongoing semi-structured individual interviews with the student and teacher case study participants based on themselves as readers both within and outside of their English classroom contexts. Ultimately, the dissertation identifies themes related to text complexity - those elements inherent to the text itself as perceived by the individual reader during the reading act - and related to text difficulty - those elements situated within a contextualized environment of the reading act, including individual reader, text, classroom, tasks, peers, and teacher - to offer provisional conclusions with the intent of reconceptualizing Rosenblatt's transactional zone toward a stronger theory of how adolescents read literary texts.

Reading in a second language (L2) is considered a necessary skill in increasingly globalized societies. Not only is reading for purposes of comprehension necessary for survival, also reading in an L2 is an important means by which L2 acquisition occurs, particularly where vocabulary is concerned. Consequently, there is a strong demand for L2 research to investigate the instructional conditions that will best promote reading comprehension and vocabulary through efficient and effective reading strategies. The current study addressed this dual need, reading comprehension and vocabulary acquisition, through an investigation of a particular type of pedagogical intervention, repeated reading (RR; i.e., multiple encounters with the same text), with high school English language learners in Taiwan. The study examined the effects of three conditions – Unassisted RR (repeated reading only), Assisted RR (repeated reading plus listening), and Control – on the participants’ reading comprehension and incidental vocabulary acquisition through a pre-test, post-test, and delayed post-test design. The results of the data from 42 participants suggested that L2 RR did not promote reading comprehension, nor did it contribute to a transfer of practice effect to new text in terms of reading comprehension. However, there were statistically significant incidental vocabulary gains and retention for the Unassisted RR group and some vocabulary gains for the Assisted RR group. When the percentage of unknown words of a text reached 10% and the participants were not provided with additional support, five repeated encounters with the text (over eight treatment sessions) were found to be inadequate in promoting reading comprehension. Nonetheless, the participants provided with such challenging condition still benefited from the incidental vocabulary acquisition. Findings may imply that a certain threshold of proficiency (e.g., percentage of known words of a text) is necessary for the beneficial effects of repeated reading to support comprehension. Additionally, RR alone may still be insufficient and additional support to RR may still be necessary for L2 learners dealing with difficult texts.

Esta pesquisa tem como objetivo analisar provas de vestibular de espanhol, disponibilizadas por algumas Instituições de Ensino Superior (IES), referentes aos processos seletivos de 2004 a 2006. Por meio de consulta às páginas web do INEP, tivemos acesso aos endereços eletrônicos das 1.130 IES da região sudeste. Após entrar em contato com todas elas, obtivemos 23 exames fornecidos por 21 instituições - duas das quais apresentavam a prova de língua espanhola na 1ª e na 2ª fase do processo seletivo. São esses os exames que integram nosso corpus de pesquisa. À luz das concepções teóricas sobre o processo de compreensão textual apresentadas por Kleiman (2004), Kato (1999) e Solé (1998), entre outros pesquisadores, e considerando os pressupostos a respeito da leitura constantes nos documentos oficiais brasileiros para o Ensino Médio, a saber: PCNEM (1999), PCN+ (2002), PCN em debate (2004) e OCNEM (2006), procedemos à análise do material coletado. Identificamos o conhecimento de mundo exigido dos candidatos em tais exames e o gênero de cada um dos 42 textos relacionados. Analisamos as 205 questões organizando-as em quatro grupos que envolveram questões de: 1. compreensão global ou parcial do texto; 2. relações entre os elementos lingüísticos do texto; 3. de conhecimento gramatical; 4. de conhecimento lexical. Por fim, analisamos a presença, ou não, de dois aspectos específicos, quais sejam, contextualização e interdisciplinaridade, presentes em todos os documentos de orientação para o Ensino Médio, organizados pelo MEC. Ainda que a maior parte das questões tenham feito referência a um texto, muitas delas ainda exigiram apenas a identificação de elementos expressos no texto. Outras, em menor proporção, continuaram avaliando apenas e/ou prioritariamente o domínio de regras gramaticais e do léxico, como se o ensino de uma língua pudesse ser reduzido a esses dois temas. Concluímos nossa pesquisa apresentando uma síntese e reflexão dos principais dados levantados sobre os exames vestibulares em língua espanhola, sobretudo quanto aos aspectos privilegiados nas provas analisadas. Por fim, entre os anexos, incluímos a relação nominal das IES contatadas, com a indicação da cidade onde se situa e o correspondente endereço eletrônico. / This research has the purpose to analyze Spanish tests of university admission exams available by some Brazilian colleges about the selective process of students in 2004 to 2006. Through consultation to the pages web of INEP we had access to the e-mails of 1,130 colleges of the south east region. After the contact with all the institutions, we got 23 tests provided by 21 institutions - two of them showed Spanish test in the first and the second stage of the selective process. These tests were organized in the corpus of this research. According to the theoretical conception about the process of textual comprehension presented by Kleiman (2004), Kato (1999) and Solé (1998), among other researchers, and considering the presupposes about the reading unchangeable in the Brazilian documents to high school, as: PCNEM (1999), PCN+ (2002), PCN em debate (2004) and OCNEM (2006), we procedured to analyze of the collected material. We identified the world knowledge demanded from the candidates in these tests and the textual gender of each one of the 42 texts presented in the corpus of this research. We analyzed 205 questions organized in 4 groups that included questions of: 1. global or partial comprehension of the text; 2. relations among linguistic elements of the text; 3. grammatical knowledge; 4. lexical knowledge. Finally, we analyzed the presence, or lack, of two specific aspects, just like, contextualization and intersubject, present in all documents of orientation to the high school, organized by MEC. Although most of the questions were connected with the text, yet many of them demanded only the identification of the elements expressed in the text. Others, in lesser proportion, continued assessing just and/or priorily the masterfulness of grammatical rules and lexical knowledge as if the teaching of a language could be reduced to these two topics. In the conclusion of this research we showed a synthesis and a reflection of the principal data checked about the tests of university admission exam in Spanish language, especially to connect with the aspects privileged in the tests that we analyzed. Finally, among annex, we included the nominal relation of the colleges contacted with the indicator of the city where they locate and their corresponding e-mail.

羅嘉怡. / 論文(哲學碩士)--香港中文大學, 1998. / 參考文獻: leaves 71-79. / 中英文摘要. / Luo Jiayi. / 表目次 --- p.iii / Chapter 第一章 --- 引言 / Chapter 一 --- 研究動機 --- p.1 / Chapter 二 --- 研究目的 --- p.12 / Chapter 三 --- 研究意義 --- p.13 / Chapter 第二章 --- 文獻回顧 / Chapter 一 --- 目標取向 --- p.16 / Chapter 二 --- 自我效能 --- p.20 / Chapter 三 --- 自我調控的學習策略 --- p.24 / Chapter 第三章 --- 研究設計 / Chapter 一 --- 研究對象與取樣方法 --- p.33 / Chapter 二 --- 研究工具 --- p.35 / Chapter 三 --- 資料搜集過程 --- p.38 / Chapter 四. --- 研究假設 --- p.40 / Chapter 五. --- 重要名詞演譯 --- p.41 / Chapter 六. --- 研究限制 --- p.43 / Chapter 第四章 --- 研究結果 / Chapter 一 --- 各量表的信度分析 --- p.44 / Chapter 二 --- 研究對象的閱讀理解成绩及各量表的平均數、標準差 --- p.46 / Chapter 三 --- 閱讀理解成績及各變項的關係 --- p.49 / Chapter 四 --- 各變項對學生的閱讀理解測驗成績的預測力 --- p.53 / Chapter 第五章 --- 研究結論討論與建議 / Chapter 一 --- 研究發現 --- p.59 / Chapter 二 --- 討論 --- p.63 / Chapter 三 --- 建議 --- p.68 / 參考文獻 / 英文參考文獻 --- p.71 / 中文參考文獻 --- p.11 / 附件一 --- p.80 / 附件二 --- p.89

I conducted 2 experiments to analyze the effects of a reading curriculum, Corrective Reading, which has a sequence that trains derived relations, on the emission of (a) derived relations defined as combinatorial entailment in Relational Frame Theory and (b) metaphors with first grade students. In Experiment 1, I compared the curriculum, which has the sequence to train derived relations to a well-known reading curriculum, RAZ Kids. RAZ Kids served as the content control. I used an experimental group design with a simultaneous treatment and a crossover feature. I selected 14 participants, who were matched then randomly assigned into 2 groups of 7. Both groups received matched instructional trials either in Corrective Reading or RAZ Kids condition, and each group was post-tested. Upon completion of the Post intervention 1 probes, each group was placed in an alternative condition, where Group 1 received the content control intervention, and Group 2 received instruction from the curriculum that has the sequence to train derived relations. Both groups increased in number of correct responses following the Corrective Reading intervention. Two kinds of analyses were done, small group and individual. In Experiment 2, I replicated Experiment I using a delayed multiple probe design across 2 first-grade dyads without a content control curriculum. I tested the effects of 5 lessons of the curriculum that has the sequence to train derived relations on the same dependent measures with an addition of implicit/explicit reading comprehension probes. The results showed that the curriculum sequence found within Corrective Reading was effective in increasing the number of correct derived relation responses, while also improving reading comprehension responses.

In three experiments, I tested the effects of the conditioned reinforcement for reading (R+Reading) on reading comprehension with 5th graders. In Experiment 1, I conducted a series of statistical analyses with data from 18 participants for one year. I administered 4 pre/post measurements for reading repertoires which included: 1) state-wide assessments, 2) district-wide assessments, 3) 20 min observational probes, and 4) preference probes. I utilized the standardized testing measurements to establish grade-level reading repertoires, while the additional two probes measured the reinforcement value of reading. Observational data were recorded in 10s whole-intervals; participants who were observed to read for 96 of the 120 intervals (80%) were considered to have R+Reading. The results demonstrated that R+Reading is significantly correlated with reading assessment outcomes. In Experiment 2, I implemented a two-year cross-sectional design with 33 participants, where I expanded the previous research to include probe trials for conditioned seeing (CS) and derivational responding (DR). Results of Experiment 2 indicated that increases in standardized testing scores were significantly correlated with R+Reading, and that CS and DR were prerequisite repertoires for the acquisition of R+Reading. In Experiment 3, I tested the effects of the peer-yoked contingency procedure on the reinforcement value of reading and assessed if increases in the reinforcement value of reading functioned to increase reading comprehension. Results indicated that increases in the reinforcement value of reading also was related to increases in reading comprehension.

No description available.