N-best error simulation for training spoken dialogue systems

A recent trend in spoken dialogue research is the use of reinforcement learning to train dialogue systems in a simulated environment. Past researchers have shown that the types of errors that are simulated can have a significant effect on simulated dialogue performance. Since modern systems typically receive an N-best list of possible user utterances, it is important to be able to simulate a full N-best list of hypotheses. This paper presents a new method for simulating such errors based on logistic regression, as well as a new method for simulating the structure of N-best lists of semantics and their probabilities, based on the Dirichlet distribution. Off-line evaluations show that the new Dirichlet model results in a much closer match to the receiver operating characteristics (ROC) of the live data. Experiments also show that the logistic model gives confusions that are closer to the type of confusions observed in live situations. The hope is that these new error models will be able to improve the resulting performance of trained dialogue systems. © 2012 IEEE.

Model reductions for inference: generality of pairwise, binary, and planar factor graphs.

We offer a solution to the problem of efficiently translating algorithms between different types of discrete statistical model. We investigate the expressive power of three classes of model-those with binary variables, with pairwise factors, and with planar topology-as well as their four intersections. We formalize a notion of "simple reduction" for the problem of inferring marginal probabilities and consider whether it is possible to "simply reduce" marginal inference from general discrete factor graphs to factor graphs in each of these seven subclasses. We characterize the reducibility of each class, showing in particular that the class of binary pairwise factor graphs is able to simply reduce only positive models. We also exhibit a continuous "spectral reduction" based on polynomial interpolation, which overcomes this limitation. Experiments assess the performance of standard approximate inference algorithms on the outputs of our reductions.

The role of human orbitofrontal cortex in value comparison for incommensurable objects.

The human orbitofrontal cortex is strongly implicated in appetitive valuation. Whether its role extends to support comparative valuation necessary to explain probabilistic choice patterns for incommensurable goods is unknown. Using a binary choice paradigm, we derived the subjective values of different bundles of goods, under conditions of both gain and loss. We demonstrate that orbitofrontal activation reflects the difference in subjective value between available options, an effect evident across valuation for both gains and losses. In contrast, activation in dorsal striatum and supplementary motor areas reflects subjects' choice probabilities. These findings indicate that orbitofrontal cortex plays a pivotal role in valuation for incommensurable goods, a critical component process in human decision making.

Neural activity associated with the passive prediction of ambiguity and risk for aversive events.

In economic decision making, outcomes are described in terms of risk (uncertain outcomes with certain probabilities) and ambiguity (uncertain outcomes with uncertain probabilities). Humans are more averse to ambiguity than to risk, with a distinct neural system suggested as mediating this effect. However, there has been no clear disambiguation of activity related to decisions themselves from perceptual processing of ambiguity. In a functional magnetic resonance imaging (fMRI) experiment, we contrasted ambiguity, defined as a lack of information about outcome probabilities, to risk, where outcome probabilities are known, or ignorance, where outcomes are completely unknown and unknowable. We modified previously learned pavlovian CS+ stimuli such that they became an ambiguous cue and contrasted evoked brain activity both with an unmodified predictive CS+ (risky cue), and a cue that conveyed no information about outcome probabilities (ignorance cue). Compared with risk, ambiguous cues elicited activity in posterior inferior frontal gyrus and posterior parietal cortex during outcome anticipation. Furthermore, a similar set of regions was activated when ambiguous cues were compared with ignorance cues. Thus, regions previously shown to be engaged by decisions about ambiguous rewarding outcomes are also engaged by ambiguous outcome prediction in the context of aversive outcomes. Moreover, activation in these regions was seen even when no actual decision is made. Our findings suggest that these regions subserve a general function of contextual analysis when search for hidden information during outcome anticipation is both necessary and meaningful.

Healthcare human reliability analysis - By heart

As part of the investigations into a surgical incident involving the accidental retention inside a patient's venous system of a guide wire for central venous catheterisation (CVC), the Human Error Assessment and Reduction Technique (HEART) was used to examine the potential for further occurrences. It was found to be time-efficient and to yield plausible probabilities of human error, although its use in healthcare has challenges, suggesting adaptation would be beneficial.

Predicting the acoustics of squeak and rattle

This paper discusses the development of a computationally efficient numerical method for predicting the acoustics of rattle events upfront in the design cycle. The method combines Finite Elements, Boundary Elements and SEA and enables the loudness of a large number of rattle events to be efficiently predicted across a broad frequency range. A low frequency random vibro-acoustic model is used in conjunction with various closed form analytical expressions in order to quickly predict impact probabilities and locations. An existing method has been extended to estimate the statistics of the contact forces across a broad frequency range. Finally, broadband acoustic radiation is predicted using standard low, mid and high frequency vibro-acoustic methods and used to estimate impact loudness. The approach is discussed and a number of validation examples are presented.

N-gram posterior probability confidence measures for statistical machine translation: An empirical study

We report an empirical study of n-gram posterior probability confidence measures for statistical machine translation (SMT). We first describe an efficient and practical algorithm for rapidly computing n-gram posterior probabilities from large translation word lattices. These probabilities are shown to be a good predictor of whether or not the n-gram is found in human reference translations, motivating their use as a confidence measure for SMT. Comprehensive n-gram precision and word coverage measurements are presented for a variety of different language pairs, domains and conditions. We analyze the effect on reference precision of using single or multiple references, and compare the precision of posteriors computed from k-best lists to those computed over the full evidence space of the lattice. We also demonstrate improved confidence by combining multiple lattices in a multi-source translation framework. © 2012 The Author(s).

Paraphrastic language models

In natural languages multiple word sequences can represent the same underlying meaning. Only modelling the observed surface word sequence can result in poor context coverage, for example, when using n-gram language models (LM). To handle this issue, this paper presents a novel form of language model, the paraphrastic LM. A phrase level transduction model that is statistically learned from standard text data is used to generate paraphrase variants. LM probabilities are then estimated by maximizing their marginal probability. Significant error rate reductions of 0.5%-0.6% absolute were obtained on a state-ofthe-art conversational telephone speech recognition task using a paraphrastic multi-level LM modelling both word and phrase sequences.

The closed thorium-transuranic fuel cycle in reduced-moderation PWRs and BWRs

Multiple recycle of long-lived actinides has the potential to greatly reduce the required storage time for spent nuclear fuel or high level nuclear waste. This is generally thought to require fast reactors as most transuranic (TRU) isotopes have low fission probabilities in thermal reactors. Reduced-moderation LWRs are a potential alternative to fast reactors with reduced time to deployment as they are based on commercially mature LWR technology. Thorium (Th) fuel is neutronically advantageous for TRU multiple recycle in LWRs due to a large improvement in the void coefficient. If Th fuel is used in reduced-moderation LWRs, it appears neutronically feasible to achieve full actinide recycle while burning an external supply of TRU, with related potential improvements in waste management and fuel utilization. In this paper, the fuel cycle of TRU-bearing Th fuel is analysed for reduced-moderation PWRs and BWRs (RMPWRs and RBWRs). RMPWRs have the advantage of relatively rapid implementation and intrinsically low conversion ratios, which is desirable to maximize the TRU burning rate. However, it is challenging to simultaneously satisfy operational and fuel cycle constraints. An RBWR may potentially take longer to implement than an RMPWR due to more extensive changes from current BWR technology. However, the harder neutron spectrum can lead to favourable fuel cycle performance. A two-stage TRU burning cycle, where the first stage is Th-Pu MOX in a conventional PWR feeding a second stage continuous burn in RMPWR or RBWR, is technically reasonable, although it is more suitable for the RBWR implementation. In this case, the fuel cycle performance is relatively insensitive to the discharge burn-up of the first stage. © 2013 Elsevier Ltd. All rights reserved.