Rebecca Bates

Dialogue act modeling for automatic tagging and recognition of conversational speech

We describe a statistical approach for modeling dialogue acts in conversational speech, i.e., speech-act-like units such as STATEMENT, Question, BACKCHANNEL, Agreement, Disagreement, and Apology. Our model detects and predicts dialogue acts based on lexical, collocational, and prosodic cues, as well as on the discourse coherence of the dialogue act sequence. The dialogue model is based on treating the discourse structure of a conversation as a hidden Markov model and the individual dialogue acts as observations emanating from the model states. Constraints on the likely sequence of dialogue acts are modeled via a dialogue act n-gram. The statistical dialogue grammar is combined with word n-grams, decision trees, and neural networks modeling the idiosyncratic lexical and prosodic manifestations of each dialogue act. We develop a probabilistic integration of speech recognition with dialogue modeling, to improve both speech recognition and dialogue act classification accuracy. Models are trained and evaluated using a large hand-labeled database of 1,155 conversations from the Switchboard corpus of spontaneous human-to-human telephone speech. We achieved good dialogue act labeling accuracy (65 based on errorful, automatically recognized words and prosody, and 71 based on word transcripts, compared to a chance baseline accuracy of 35 and human accuracy of 84) and a small reduction in word recognition error.

Direct regulation of the Xenopus engrailed-2 promoter by the Wnt signaling pathway, and a molecular screen for Wnt-responsive genes, confirm a role for Wnt signaling during neural patterning in Xenopus

The co-activation of Wnt signaling and concomitant inhibition of BMP signaling has previously been implicated in vertebrate neural patterning, as evidenced by the combinatorial induction of engrailed-2 and krox-20 in Xenopus. However, screens have not previously been conducted to identify additional potential target genes. Using a PCR-based screening method we determined that XA-1, xCRISP, UVS.2, two UVS.2-related genes, and xONR1 are induced in response to Xwnt-3a and a BMP-antagonist, noggin. Two additional genes, connexin 30 and retinoic acid receptor gamma were induced by Xwnt-3a alone. To determine whether any of the induced genes are direct targets of Wnt signaling, we focussed on engrailed-2. In the present study we show that the Xenopus engrailed-2 promoter contains three consensus binding sites for LEF/TCF, which are HMG box transcription factors which bind to beta-catenin in response to activation of the Wnt- 1 signaling pathway. An engrailed-2 promoter luciferase reporter construct containing these LEF/TCF sites is induced in embryo explant assays by the combination of Xwnt-3a or beta-catenin and noggin. These LEF/TCF sites are required for expression of engrailed-2, as a dominant negative Xtcf-3 blocks expression of endogenous engrailed-2 as well as expression of the reporter construct. Moreover, mutation of these three LEF/TCF sites abrogates expression of the reporter construct in response to noggin and Xwnt-3a or beta-catenin. We conclude that the engrailed-2 gene is a direct target of the Wnt signaling pathway, and that Wnt signaling works with BMP antagonists to regulate gene expression during patterning of the developing nervous system of Xenopus.

Symbolic phonetic features for modeling of pronunciation variation

A significant source of variation in spontaneous speech is due to intra-speaker pronunciation changes, often realized as small feature changes, e.g., nasalized vowels or affricated stops, rather than full phone transformations. Previous computational modeling of pronunciation variation has typically involved transformations from one phone to another, in part because most speech processing systems use phone-based units. Here, a phonetic-feature-based prediction model is presented where phones are represented by a vector of symbolic features that can be on, off, unspecified or unused. Feature interaction is examined using different groupings of possibly dependent features, and a hierarchical grouping with conditional dependencies led to the best results. Feature-based models are shown to be more efficient than phone-based models, in the sense of requiring fewer parameters to predict variation while giving smaller distance and perplexity values when comparing predictions to the hand-labeled reference. A parsimonious model is better suited to incorporating new conditioning factors, and this work investigates high-level information sources, including both text (syntax, discourse) and prosody cues. Experiments show that feature-based models benefit from prosody cues, but not text, and that phone-based models do not benefit from any of the high-level cues explored here.

Science fiction in computer science education

The use of science fiction (SF) to engage students in computer science learning is becoming more popular [1-6]. There is ample material available to help both undergraduate and graduate students make connections between technical content and human experience, from Star Trek to The Hitchhikers Guide to the Galaxy to 2001: A Space Odyssey to I, Robot and many others. Fiction can be included in technical courses or used to draw students into the field in introductory classes. The panelists, who represent a range of schools, perspectives and classes, will present brief overviews (5-8 minutes) of how they have used science fiction to engage students in technical topics as well as ethical and societal issues related to computing. After the overviews, there will be plenty of time for discussion of examples used within the community and ways to make connections between science fiction and particular classes or topics. We will be gathering additional examples from the discussion and making them available online.

Awareness of and receptiveness to active learning strategies among STEM faculty

Despite strong evidence of the positive impact of active learning strategies, STEM faculty demonstrate a spectrum of receptiveness to incorporating active learning into their classrooms, and for a variety of reasons, engineering classes continue to be dominated by a passive lecture style. This paper draws on data from a four-year study that investigated the use of five social instruction strategies, including active learning. Twenty-four STEM faculty at 4 institutions were interviewed regarding their understanding of and attitudes toward these strategies. This paper focuses on the results of the active learning component of these interviews. Faculty most often interpreted active learning as what students do and viewed self-motivation as a key component of what students think while active learning. These results, while drawn from a small sample population, can nevertheless make an important contribution to understanding why passive learning remains predominant in the STEM classroom. This paper examines how the findings from this study can inform efforts to promote changes in STEM education that would bring more active learning to the classroom.

Special session: Agents for STEM change — Articulating the goals of our community

Engineering and Computer Science (E&CS) Education is an emerging discipline with a brief history and the unfortunate particularity that many folks outside our discipline are confused as to our purpose. In this special session, we will use two case studies to frame the larger questions around E&CS Education goals and help draw the conversation from practice to philosophy while creating a safe space for open conversation.

Experiences with Shared Resources for Research and Education in Speech and Language Processing

Resource barriers can prevent capable researchers from participating in the speech and language community and can make it difficult to support learning and participation in our field at a wide variety of institutions. Sharing resources, whether software, processed data, experimental methodologies or virtual machines, can reduce the barrier to entry and potentially broaden participation in speech and language research and improve workforce development. As an introduction to the special session on Sharing Research and Education Resources for Understanding Speech Processing, we outline current trends and requirements for expanding participation in speech processing research. A qualitative research approach was used. Faculty at a variety of institutions have been interviewed and have participated in reflection writing about needs, tools, challenges, and successes. Themes from reflections were generated using a grounded theory approach and were used to code interviews for related evidence. This paper describes the educational and research challenges experienced by faculty as users of resources, rather than the details of specific resources provided. The goal is to engage in a stronger dialog between users and providers so that needs and resources are better aligned. A case study of a shared resource used at several universities highlights this dialog.

Special session: Agents for change in engineering & computer science education

Engineering and Computer Science (E&CS) Education is an emerging discipline and is a subset of the larger field of STEM (science, technology, engineering and math) education. E&CS Education has a relatively brief history, and many individuals who are not directly involved in the discipline are often confused about its purpose. In this special session, we will use up to two background stories to frame the larger questions around E&CS Education goals and help draw the conversation from practice to philosophy while creating a safe space for open conversation addressing issues of change in our discipline.

Credentialing in the CSET education change process

Efforts have been made to improve technical and professional skills in engineering graduates, but little widespread change in pedagogy has occurred within U.S. engineering education institutions. Our group studied the genesis and implementation of an innovative engineering curriculum (Iron Range Engineering) through a series of interviews with a wide range of stakeholders. Using a grounded theory approach, we found that to “shoehorn” an innovative curriculum into a traditional university setting required ad hoc solutions - almost akin to hacking a system. The findings in the study of this process also showed that the most common barriers to widespread educational innovation can be framed as credentialing issues, whether as excuses for not implementing change or as legitimate obstacles. At the root of the credentialing issue is the ubiquitous standard unit of effort-the credit hour, which was originally designed simply to measure faculty workload rather than student learning. This paper seeks to describe the breadth of credentialing in terms of scope and groups involved. Finally, we propose conversations that change agents in CSET education can use to turn credentialing into a lever for systemic curricular transformation.

Evaluation and continuous improvement in a multidisciplinary S-STEM program focusing on professional skills, goals & mentoring

Four years of student feedback from participants in a sponsored scholarship program is evaluated to assess perceived benefit of program activities that go beyond awarding scholarships. The program is structured to help students develop professional skills such as communication, teamwork, job readiness, and goal setting in a context that allows for formal and informal mentoring. Annual themes often related to current issues affecting society provide a setting to motivate these skills. Results show that students value skills and experiences outside of their major requirements. This work describes the program and its process for continuous improvement and provides insights into student willingness to participate in personal and professional activities on top of rigorous STEM major requirements.