• Evidence from the research literature indicates that both audience response systems (ARS) and guided inquiry worksheets (GIW) can lead to greater student engagement, learning, and equity in the STEM classroom. We compare the use of these two tools in large enrollment STEM courses delivered in different contexts, one in biology and one in engineering. The instructors studied utilized each of the active learning tools differently. In the biology course, ARS questions were used mainly to check in with students and assess if they were correctly interpreting and understanding worksheet questions. The engineering course presented ARS questions that afforded students the opportunity to apply learned concepts to new scenarios towards improving students conceptual understanding. In the biology course, the GIWs were primarily used in stand-alone activities, and most of the information necessary for students to answer the questions was contained within the worksheet in a context that aligned with a disciplinary model. In the engineering course, the instructor intended for students to reference their lecture notes and rely on their conceptual knowledge of fundamental principles from the previous ARS class session in order to successfully answer the GIW questions. However, while their specific implementation structures and practices differed, both instructors used these tools to build towards the same basic disciplinary thinking and sense-making processes of conceptual reasoning, quantitative reasoning, and metacognitive thinking.
  • Educational research has shown that narratives are useful tools that can help young students make sense of scientific phenomena. Based on previous research, I argue that narratives can also become tools for high school students to make sense of concepts such as the electric field. In this paper I examine high school students visual and oral narratives in which they describe the interaction among electric charges as if they were characters of a cartoon series. The study investigates: given the prompt to produce narratives for electrostatic phenomena during a classroom activity prior to receiving formal instruction, (1) what ideas of electrostatics do students attend to in their narratives?; (2) what role do students narratives play in their understanding of electrostatics? The participants were a group of high school students engaged in an open-ended classroom activity prior to receiving formal instruction about electrostatics. During the activity, the group was asked to draw comic strips for electric charges. In addition to individual work, students shared their work within small groups as well as with the whole group. Post activity, six students from a small group were interviewed individually about their work. In this paper I present two cases in which students produced narratives to express their ideas about electrostatics in different ways. In each case, I present student work for the comic strip activity (visual narratives), their oral descriptions of their work (oral narratives) during the interview and/or to their peers during class, and the their ideas of the electric interactions expressed through their narratives.
  • This study investigates a group of high school students in a physics classroom interacting with a computer simulation that simulates the electrostatic interaction as a hockey game, the Electric Field Hockey. The activity featured in this study took place prior to the students receiving formal instruction about the electric field. The learning goal was to allow students to explore the simulated electrostatic phenomena. The study asks the following research question: How do high school students collectively explore simulated electrostatic phenomena while interacting with the Electric Field Hockey computer simulation during a classroom activity? In this paper, through careful analysis of classroom videos, I present a case study about the group mentioned above. The results show that high school students encountered and dealt with multiple types of tensions in the activity and explored the electrostatic phenomena simulated by the Electric Field Hockey. Students prioritized exploring different types of phenomena when they encountered different types of tensions. Based on the findings I propose: When carrying out an activity with a novel educational tool, a teacher can attend to, and jump in at appropriate points, when tensions arises thus to orient students to explore the simulated physical phenomena in more productive ways.
  • This study investigates the representations and understandings of electric fields expressed by Chinese high school students 15 to 16 years old who have not received high school level physics instruction. The physics education research literature has reported students conceptions of electric fields post-instruction as indicated by students performance on textbook-style questions. It has, however, inadequately captured student ideas expressed in other situations yet informative to educational research. In this study, we explore students ideas of electric fields pre-instruction as shown by students representations produced in open-ended activities. 92 participant students completed a worksheet that involved drawing comic strips about electric charges as characters of a cartoon series. Three students who had spontaneously produced arrow diagrams were interviewed individually after class. We identified nine ideas related to electric fields that these three students spontaneously leveraged in the comic strip activity. In this paper, we describe in detail each idea and its situated context. As most research in the literature has understood students as having relatively fixed conceptions and mostly identified divergences in those conceptions from canonical targets, this study shows students reasoning to be more variable in particular moments, and that variability includes common sense resources that can be productive for learning about electric fields.
  • We present Deep Speaker, a neural speaker embedding system that maps utterances to a hypersphere where speaker similarity is measured by cosine similarity. The embeddings generated by Deep Speaker can be used for many tasks, including speaker identification, verification, and clustering. We experiment with ResCNN and GRU architectures to extract the acoustic features, then mean pool to produce utterance-level speaker embeddings, and train using triplet loss based on cosine similarity. Experiments on three distinct datasets suggest that Deep Speaker outperforms a DNN-based i-vector baseline. For example, Deep Speaker reduces the verification equal error rate by 50% (relatively) and improves the identification accuracy by 60% (relatively) on a text-independent dataset. We also present results that suggest adapting from a model trained with Mandarin can improve accuracy for English speaker recognition.
  • While question answering (QA) with neural network, i.e. neural QA, has achieved promising results in recent years, lacking of large scale real-word QA dataset is still a challenge for developing and evaluating neural QA system. To alleviate this problem, we propose a large scale human annotated real-world QA dataset WebQA with more than 42k questions and 556k evidences. As existing neural QA methods resolve QA either as sequence generation or classification/ranking problem, they face challenges of expensive softmax computation, unseen answers handling or separate candidate answer generation component. In this work, we cast neural QA as a sequence labeling problem and propose an end-to-end sequence labeling model, which overcomes all the above challenges. Experimental results on WebQA show that our model outperforms the baselines significantly with an F1 score of 74.69% with word-based input, and the performance drops only 3.72 F1 points with more challenging character-based input.
  • Neural machine translation (NMT) aims at solving machine translation (MT) problems using neural networks and has exhibited promising results in recent years. However, most of the existing NMT models are shallow and there is still a performance gap between a single NMT model and the best conventional MT system. In this work, we introduce a new type of linear connections, named fast-forward connections, based on deep Long Short-Term Memory (LSTM) networks, and an interleaved bi-directional architecture for stacking the LSTM layers. Fast-forward connections play an essential role in propagating the gradients and building a deep topology of depth 16. On the WMT'14 English-to-French task, we achieve BLEU=37.7 with a single attention model, which outperforms the corresponding single shallow model by 6.2 BLEU points. This is the first time that a single NMT model achieves state-of-the-art performance and outperforms the best conventional model by 0.7 BLEU points. We can still achieve BLEU=36.3 even without using an attention mechanism. After special handling of unknown words and model ensembling, we obtain the best score reported to date on this task with BLEU=40.4. Our models are also validated on the more difficult WMT'14 English-to-German task.