Discriminative spoken language understanding using word confusion networks

Current commercial dialogue systems typically use hand-crafted grammars for Spoken Language Understanding (SLU) operating on the top one or two hypotheses output by the speech recogniser. These systems are expensive to develop and they suffer from significant degradation in performance when faced with recognition errors. This paper presents a robust method for SLU based on features extracted from the full posterior distribution of recognition hypotheses encoded in the form of word confusion networks. Following [1], the system uses SVM classifiers operating on n-gram features, trained on unaligned input/output pairs. Performance is evaluated on both an off-line corpus and on-line in a live user trial. It is shown that a statistical discriminative approach to SLU operating on the full posterior ASR output distribution can substantially improve performance both in terms of accuracy and overall dialogue reward. Furthermore, additional gains can be obtained by incorporating features from the previous system output. © 2012 IEEE.

Language model cross adaptation for LVCSR system combination

State-of-the-art large vocabulary continuous speech recognition (LVCSR) systems often combine outputs from multiple sub-systems that may even be developed at different sites. Cross system adaptation, in which model adaptation is performed using the outputs from another sub-system, can be used as an alternative to hypothesis level combination schemes such as ROVER. Normally cross adaptation is only performed on the acoustic models. However, there are many other levels in LVCSR systems' modelling hierarchy where complimentary features may be exploited, for example, the sub-word and the word level, to further improve cross adaptation based system combination. It is thus interesting to also cross adapt language models (LMs) to capture these additional useful features. In this paper cross adaptation is applied to three forms of language models, a multi-level LM that models both syllable and word sequences, a word level neural network LM, and the linear combination of the two. Significant error rate reductions of 4.0-7.1% relative were obtained over ROVER and acoustic model only cross adaptation when combining a range of Chinese LVCSR sub-systems used in the 2010 and 2011 DARPA GALE evaluations. © 2012 Elsevier Ltd. All rights reserved.

Probabilistic soil identification based on cone penetration tests

In geotechnical engineering, soil classification is an essential component in the design process. Field methods such as the cone penetration test (CPT) can be used as less expensive and faster alternatives to sample retrieval and testing. Unfortunately, current soil classification charts based on CPT data and laboratory measurements are too generic, and may not provide an accurate prediction of the soil type. A probabilistic approach is proposed here to update and modify soil identification charts based on site-specific CPT data. The probability that a soil is correctly classified is also estimated. The updated identification chart can be used for a more accurate prediction of the classification of the soil, and can account for prior information available before conducting the tests, site-specific data, and measurement errors. As an illustration, the proposed approach is implemented using CPT data from the Treporti Test Site (TTS) near Venice (Italy) and the National Geotechnical Experimentation Sites (NGES) at Texas A&M University. The applicability of the site-specific chart for other sites in Venice Lagoon is assessed using data from the Malamocco test site, approximately 20 km from TTS.

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.

Friction-induced vibration: Model development and comparison with large-scale experimental tests

This paper presents a comparison between theoretical predictions and experimental results from a pin-on-disc test rig exploring friction-induced vibration. The model is based on a linear stability analysis of two systems coupled by sliding contact at a single point. Predictions are compared with a large volume of measured squeal initiations that have been post-processed to extract growth rates and frequencies at the onset of squeal. Initial tests reveal the importance of including both finite contact stiffness and a velocity-dependent dynamic model for friction, giving predictions that accounted for nearly all major clusters of squeal initiations from 0 to 5 kHz. However, a large number of initiations occurred at disc mode frequencies that were not predicted with the same parameters. These frequencies proved remarkably difficult to destabilise, requiring an implausibly high coefficient of friction. An attempt has been made to estimate the dynamic friction behaviour directly from the squeal initiation data, revealing complex-valued frequency-dependent parameters for a new model of linearised dynamic friction. These new parameters readily destabilised the disc modes and provided a consistent model that could account for virtually all initiations from 0 to 15 kHz. The results suggest that instability thresholds for a wide range of squeal-type behaviour can be predicted, but they highlight the central importance of a correct understanding and accurate description of dynamic friction at the sliding interface. © 2013 Elsevier Ltd. All rights reserved.

Tests of an adaptive QM/MM calculation on free energy profiles of chemical reactions in solution.

We present reaction free energy calculations using the adaptive buffered force mixing quantum mechanics/molecular mechanics (bf-QM/MM) method. The bf-QM/MM method combines nonadaptive electrostatic embedding QM/MM calculations with extended and reduced QM regions to calculate accurate forces on all atoms, which can be used in free energy calculation methods that require only the forces and not the energy. We calculate the free energy profiles of two reactions in aqueous solution: the nucleophilic substitution reaction of methyl chloride with a chloride anion and the deprotonation reaction of the tyrosine side chain. We validate the bf-QM/MM method against a full QM simulation, and show that it correctly reproduces both geometrical properties and free energy profiles of the QM model, while the electrostatic embedding QM/MM method using a static QM region comprising only the solute is unable to do so. The bf-QM/MM method is not explicitly dependent on the details of the QM and MM methods, so long as it is possible to compute QM forces in a small region and MM forces in the rest of the system, as in a conventional QM/MM calculation. It is simple, with only a few parameters needed to control the QM calculation sizes, and allows (but does not require) a varying and adapting QM region which is necessary for simulating solutions.

Boundary conditions in physical model tests-the influence of deck pinning on the response of piled bridge abutments in laterally spreading soils

Physical models are widely used in the study of geotechnical earthquake engineering phenomena, and the comparison of modelling results to observations from field reconnaissance provides a transparent means of evaluating the design of our physical models. This paper compares centrifuge tests of pile groups in laterally spreading slopes with the response of piled bridge abutments in the 2011 Christchurch earthquake. We show that the model foundation's fixity conditions strongly affect the success with which the mechanism of response of the real abutments is replicated in the tests. © 2012 American Society of Civil Engineers.

Paraphrastic language models and combination with neural network 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, paraphrastic LMs were proposed in previous research and successfully applied to a US English conversational telephone speech transcription task. In order to exploit the complementary characteristics of paraphrastic LMs and neural network LMs (NNLM), the combination between the two is investigated in this paper. To investigate paraphrastic LMs' generalization ability to other languages, experiments are conducted on a Mandarin Chinese broadcast speech transcription task. Using a paraphrastic multi-level LM modelling both word and phrase sequences, significant error rate reductions of 0.9% absolute (9% relative) and 0.5% absolute (5% relative) were obtained over the baseline n-gram and NNLM systems respectively, after a combination with word and phrase level NNLMs. © 2013 IEEE.

High-performance bowing machine tests of bowed-string transients

The development is described of a computer-controlled bowing machine that can bow a string with a range of gestures that match or exceed the capabilities of a human violinist. Example measurements of string vibration under controlled bowing conditions are shown, including a Schelleng diagram and a set of Guettler diagrams, for the open D string of a cello. For some results a rosin-coated rod was used in place of a conventional bow, to provide quantitative data for comparison with theoretical predictions. The results show qualitative consistency with the predictions of Schelleng and Guettler, but details are revealed that go beyond the limitations of existing models. © S. Hirzel Verlag · EAA.

Automatic construction and natural-language description of nonparametric regression models

Copyright © 2014, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. This paper presents the beginnings of an automatic statistician, focusing on regression problems. Our system explores an open-ended space of statistical models to discover a good explanation of a data set, and then produces a detailed report with figures and natural- language text. Our approach treats unknown regression functions non- parametrically using Gaussian processes, which has two important consequences. First, Gaussian processes can model functions in terms of high-level properties (e.g. smoothness, trends, periodicity, changepoints). Taken together with the compositional structure of our language of models this allows us to automatically describe functions in simple terms. Second, the use of flexible nonparametric models and a rich language for composing them in an open-ended manner also results in state- of-the-art extrapolation performance evaluated over 13 real time series data sets from various domains.