Task-oriented dialog presents a difficult challenge encompassing multiple problems including mult... more Task-oriented dialog presents a difficult challenge encompassing multiple problems including multi-turn language understanding and generation, knowledge retrieval and reasoning, and action prediction. Modern dialog systems typically begin by converting conversation history to a symbolic object referred to as belief state by using supervised learning. The belief state is then used to reason on an external knowledge source whose result along with the conversation history is used in action prediction and response generation tasks independently. Such a pipeline of individually optimized components not only makes the development process cumbersome but also makes it non-trivial to leverage session-level user reinforcement signals. In this paper, we develop Neural Assistant: a single neural network model that takes conversation history and an external knowledge source as input and jointly produces both text response and action to be taken by the system as output. The model learns to reason on the provided knowledge source with weak supervision signal coming from the text generation and the action prediction tasks, hence removing the need for belief state annotations. In the MultiWOZ dataset, we study the effect of distant supervision, and the size of knowledge base on model performance. We find that the Neural Assistant without belief states is able to incorporate external knowledge information achieving higher factual accuracy scores compared to Transformer. In settings comparable to reported baseline systems, Neural Assistant when provided with oracle belief state significantly improves language generation performance. * Equal contribution † Work done when all authors were at Google 3 We ignore speech-to-text and text-to-speech components in this work.
Categorical and coordinate stimulus processing were hypothesized by Kosslyn (1987) to be laterali... more Categorical and coordinate stimulus processing were hypothesized by Kosslyn (1987) to be lateralized visual tasks, differentiated by task-relevant spatial frequencies. Slotnick et al. (2001) directly tested Kosslyn’s hypothesis and concluded that the lateralization presents only when tasks are sufficiently difficult. Our differential encoding model is a three layer neural network that accounts for lateralization in visual processing via the biologically plausible mechanism of differences in connection spread of long-range lateral neural connections (Hsiao, Cipollini, & Cottrell, 2013). We show that our model accounts for Slotnick’s data and that Slotnick’s analysis does not convincingly explain their results. Instead, we propose that Kosslyn’s initial hypothesis was based on an incorrect assumption: categorical and coordinate stimuli are not solely differentiated by spatial frequencies. The results that our model captures cannot be reproduced by Ivry and Robertson’s (1998) Double Fi...
This article may be used for non-commercial purposes in accordance with Wiley Terms and Condition... more This article may be used for non-commercial purposes in accordance with Wiley Terms and Conditions for Self-Archiving." A note on versions: The version presented here may differ from the published version or, version of record, if you wish to cite this item you are advised to consult the publisher's version. Please see the 'permanent WRAP URL' above for details on accessing the published version and note that access may require a subscription.
Sixty patients with unilateral stroke (half with left hemisphere damage and half with right hemis... more Sixty patients with unilateral stroke (half with left hemisphere damage and half with right hemisphere damage) and a control group (N = 15) matched for age and educational level were tested in two experiments. In one experiment they were first shown, on each trial, a sample drawing depicting one or more objects. Following a short delay, they were asked to identify the drawing when it was paired with a drawing in which the same object(s) was transformed in categorical or coordinate spatial relations. In the other experiment, the same subjects first were shown, on each trial, a sample drawing. They then judged which of two variants (each in one type of spatial relation) looked more similar to the sample drawing. Typically, patients with left-sided stroke mistakenly identified the categorical transformation for the sample drawing in the first task; in the second task, they judged the categorical transformation as more similar to the sample drawing. Patients with right-sided stroke mist...
In this paper, we estimate the influence of social networks on educational attainment and behavio... more In this paper, we estimate the influence of social networks on educational attainment and behavioral outcomes of students in school. More specifically, we investigate how separating from pre-existing social networks during the transition from elementary to middle school affect students' academic progress and school and social satisfaction. We use social networks identified by the students themselves in elementary school, as part of a unique aspect of the Tel Aviv school application process which allows sixth-grade students to designate their middle schools of choice and to list up to eight friends with whom they wish to attend that school. The lists create natural "friendship hierarchies" that we exploit in our analysis. We designate the three categories of social networks that stem from these lists as follows: (1) reciprocal friends (students who list one another); and for those whose friendship requests did not match: (2) followers (those who listed fellow students as friends but were not listed as friends by these same fellow students) and (3) non-reciprocal friends (parallel to followers). Our identification strategy is based on a conditional random assignment model: in Tel Aviv middle schools students' are randomly assigned to classes within a given school. Therefore, conditional on the number of friends a student has at her school, the number of friends she attends class with should be random. Our results suggest that the presence of reciprocal friends and followers in class has a positive and significant effect on test scores in English, math, and Hebrew. However, the number of friends in the social network beyond the first circle of reciprocal friends has no effect at all on students. In addition, the presence of non-reciprocal friends in class has a negative effect on a student's learning outcomes. We find that these effects have interesting patterns of heterogeneity by gender, ability and age of students. In addition, we find that these various types of social networks have positive effects on other measures of non-cognitive behavioral outcomes, including social and overall happiness in school and whether one exhibits violent behavior.
Task-oriented dialog presents a difficult challenge encompassing multiple problems including mult... more Task-oriented dialog presents a difficult challenge encompassing multiple problems including multi-turn language understanding and generation, knowledge retrieval and reasoning, and action prediction. Modern dialog systems typically begin by converting conversation history to a symbolic object referred to as belief state by using supervised learning. The belief state is then used to reason on an external knowledge source whose result along with the conversation history is used in action prediction and response generation tasks independently. Such a pipeline of individually optimized components not only makes the development process cumbersome but also makes it non-trivial to leverage session-level user reinforcement signals. In this paper, we develop Neural Assistant: a single neural network model that takes conversation history and an external knowledge source as input and jointly produces both text response and action to be taken by the system as output. The model learns to reason on the provided knowledge source with weak supervision signal coming from the text generation and the action prediction tasks, hence removing the need for belief state annotations. In the MultiWOZ dataset, we study the effect of distant supervision, and the size of knowledge base on model performance. We find that the Neural Assistant without belief states is able to incorporate external knowledge information achieving higher factual accuracy scores compared to Transformer. In settings comparable to reported baseline systems, Neural Assistant when provided with oracle belief state significantly improves language generation performance. * Equal contribution † Work done when all authors were at Google 3 We ignore speech-to-text and text-to-speech components in this work.
Categorical and coordinate stimulus processing were hypothesized by Kosslyn (1987) to be laterali... more Categorical and coordinate stimulus processing were hypothesized by Kosslyn (1987) to be lateralized visual tasks, differentiated by task-relevant spatial frequencies. Slotnick et al. (2001) directly tested Kosslyn’s hypothesis and concluded that the lateralization presents only when tasks are sufficiently difficult. Our differential encoding model is a three layer neural network that accounts for lateralization in visual processing via the biologically plausible mechanism of differences in connection spread of long-range lateral neural connections (Hsiao, Cipollini, & Cottrell, 2013). We show that our model accounts for Slotnick’s data and that Slotnick’s analysis does not convincingly explain their results. Instead, we propose that Kosslyn’s initial hypothesis was based on an incorrect assumption: categorical and coordinate stimuli are not solely differentiated by spatial frequencies. The results that our model captures cannot be reproduced by Ivry and Robertson’s (1998) Double Fi...
This article may be used for non-commercial purposes in accordance with Wiley Terms and Condition... more This article may be used for non-commercial purposes in accordance with Wiley Terms and Conditions for Self-Archiving." A note on versions: The version presented here may differ from the published version or, version of record, if you wish to cite this item you are advised to consult the publisher's version. Please see the 'permanent WRAP URL' above for details on accessing the published version and note that access may require a subscription.
Sixty patients with unilateral stroke (half with left hemisphere damage and half with right hemis... more Sixty patients with unilateral stroke (half with left hemisphere damage and half with right hemisphere damage) and a control group (N = 15) matched for age and educational level were tested in two experiments. In one experiment they were first shown, on each trial, a sample drawing depicting one or more objects. Following a short delay, they were asked to identify the drawing when it was paired with a drawing in which the same object(s) was transformed in categorical or coordinate spatial relations. In the other experiment, the same subjects first were shown, on each trial, a sample drawing. They then judged which of two variants (each in one type of spatial relation) looked more similar to the sample drawing. Typically, patients with left-sided stroke mistakenly identified the categorical transformation for the sample drawing in the first task; in the second task, they judged the categorical transformation as more similar to the sample drawing. Patients with right-sided stroke mist...
In this paper, we estimate the influence of social networks on educational attainment and behavio... more In this paper, we estimate the influence of social networks on educational attainment and behavioral outcomes of students in school. More specifically, we investigate how separating from pre-existing social networks during the transition from elementary to middle school affect students' academic progress and school and social satisfaction. We use social networks identified by the students themselves in elementary school, as part of a unique aspect of the Tel Aviv school application process which allows sixth-grade students to designate their middle schools of choice and to list up to eight friends with whom they wish to attend that school. The lists create natural "friendship hierarchies" that we exploit in our analysis. We designate the three categories of social networks that stem from these lists as follows: (1) reciprocal friends (students who list one another); and for those whose friendship requests did not match: (2) followers (those who listed fellow students as friends but were not listed as friends by these same fellow students) and (3) non-reciprocal friends (parallel to followers). Our identification strategy is based on a conditional random assignment model: in Tel Aviv middle schools students' are randomly assigned to classes within a given school. Therefore, conditional on the number of friends a student has at her school, the number of friends she attends class with should be random. Our results suggest that the presence of reciprocal friends and followers in class has a positive and significant effect on test scores in English, math, and Hebrew. However, the number of friends in the social network beyond the first circle of reciprocal friends has no effect at all on students. In addition, the presence of non-reciprocal friends in class has a negative effect on a student's learning outcomes. We find that these effects have interesting patterns of heterogeneity by gender, ability and age of students. In addition, we find that these various types of social networks have positive effects on other measures of non-cognitive behavioral outcomes, including social and overall happiness in school and whether one exhibits violent behavior.
Uploads
Papers by Vishaal Prasad