Modeling under-resourced languages for speech recognition

One particular problem in large vocabulary continuous speech recognition for low-resourced languages is finding relevant training data for the statistical language models. Large amount of data is required, because models should estimate the probability for all possible word sequences. For Finnish, Estonian and the other fenno-ugric languages a special problem with the data is the huge amount of different word forms that are common in normal speech. The same problem exists also in other language technology applications such as machine translation, information retrieval, and in some extent also in other morphologically rich languages. In this paper we present methods and evaluations in four recent language modeling topics: selecting conversational data from the Internet, adapting models for foreign words, multi-domain and adapted neural network language modeling, and decoding with subword units. Our evaluations show that the same methods work in more than one language and that they scale down to smaller data resources.     

Exceptions in remote procedure calls using C++ template metaprogramming

Serialization of data is needed when information is passed among program entities that have no shared memory. For remote procedure calls, this includes serialization of exception objects in addition to more traditional parameters and return values. In C++, serialization of exceptions is more complicated than parameters and return values, since internal copying and passing of exceptions is handled differently from C++ parameters and return values. This article presents a light‐weight template metaprogramming‐based mechanism for passing C++ exceptions in remote procedure calls, remote method invocations, and other situations where caller and callee do not have a shared address space. This mechanism has been implemented and tested in the KC++ concurrent active object system. Copyright © 2006 John Wiley & Sons, Ltd.     

Neural network modelling of word production in Finnish: coding semantic and non-semantic features

The objective of our research is to computationally model word production and its disorders by means of artificial neural networks. In the current study we develop and analyze an algorithm that generates a distributed semantic coding from a given semantic tree-structure classification of words. With the algorithm it is possible to generate semantic representations that are compact and easy to modify. This renders the coding method suitable for our multilayer perceptron-based neural network model of word production. The model is shown to be able to account for a variety of performance patterns observed in four Finnish aphasia patients suffering from word-finding difficulties.     

Lippmann-Schwinger solvers for the computational homogenization of materials with pores

We show that under suitable hypotheses on the nonporous material law and a geometric regularity condition on the pore space, Moulinec-Suquet's basic solution scheme converges linearly. We also discuss for which derived solvers a (super)linear convergence behavior may be obtained, and for which such results do not hold, in general. The key technical argument relies on a specific subspace on which the homogenization problem is nondegenerate, and which is preserved by iterations of the basic scheme. Our line of argument is based in the nondiscretized setting, and we draw conclusions on the convergence behavior for discretized solution schemes in FFT-based computational homogenization. Also, we see how the geometry of the pores' interface enters the convergence estimates. We provide computational experiments underlining our claims.     

Computational homogenization of elasticity on a staggered grid

In this article, we propose to discretize the problem of linear elastic homogenization by finite differences on a staggered grid and introduce fast and robust solvers. Our method shares some properties with the FFT‐based homogenization technique of Moulinec and Suquet, which has received widespread attention recently because of its robustness and computational speed. These similarities include the use of FFT and the resulting performing solvers. The staggered grid discretization, however, offers three crucial improvements. Firstly, solutions obtained by our method are completely devoid of the spurious oscillations characterizing solutions obtained by Moulinec–Suquet's discretization. Secondly, the iteration numbers of our solvers are bounded independently of the grid size and the contrast. In particular, our solvers converge for three‐dimensional porous structures, which cannot be handled by Moulinec–Suquet's method. Thirdly, the finite difference discretization allows for algorithmic variants with lower memory consumption. More precisely, it is possible to reduce the memory consumption of the Moulinec–Suquet algorithms by 50%. We underline the effectiveness and the applicability of our methods by several numerical experiments of industrial scale. Copyright © 2015 John Wiley & Sons, Ltd.     

Polycyclic aromatic hydrocarbons (PAHs) in soot produced by combustion of polystyrene,polypropylene, and wood

The smoke and soot produced by the combustion of plastics or wood in a domestic stove or fireplace contain many poisonous compounds, including the polycyclic aromatic hydrocarbons (PAHs), many of which are carcinogenic. PAHs were selected as the subject of our study to gain a better sense of the hazards of burning plastics. Small samples of polystyrene, polypropylene, and wood underwent combustion in a tubular oven at 700°C; in addition, polystyrene and wood were combusted at room temperature. After their extraction and purification, the PAHs were analyzed by gas chromatography and mass spectrometry. Conditions in the hot oven promoted soot production, whereas combustion at room temperature led to somewhat more complete combustion. The PAH profiles of the examined materials resembled each other to some extent, though the original chemical structure of the polymeric materials varied a great deal. However, clear differences between the materials could be detected from the soot extracts, the soot of polystyrene being especially rich in compounds containing remnants of the polymer structure. Carcinogenic activity caused by the PAHs can be assumed to be of the same order of magnitude as soot from the combustion of wood.     

The effect of ageing on the histochemically demonstrable catecholamines in the hypogastric (main pelvic) ganglion of the rat

The effect of ageing on adrenergic neurons was studied in the hypogastric ganglion of the male rat using the formaldehyde-induced fluorescence method. With age, two changes were obvious. Firstly, the fluorescence intensity of the neurons decreased throughout postnatal life and the number of completely non-fluorescent adrenergic neurons increased. Secondly, the amount of non-specific fluorescence due to lipofuscin pigment increased. The pigment fluorescence was also found around the neurons in satellite cells.     

Consideration of the evolution of the Saccharomyces cerevisiae MEL gene family on the basis of the nucleotide sequences of the genes and their flanking regions

Analysis of the DNA sequences of new members of the Saccharomyces cerevisiae MEL1-MEL10 gene family showed high homology between the members. The MEL gene family, α-galactosidase-coding sequences, have diverged into two groups; one consisting of MEL1 and MEL2 and the other of MEL3-MEL10. In two S. cerevisiae strains containing five or seven MEL genes each, all the genes are nearly identical, suggesting very rapid distribution of the gene to separate chromosomes. The sequence homology and the abrupt change to sequence heterogeneity at the centromere-proximal 3′ end of the MEL genes suggest that the distribution of the genes to new chromosomal locations has occurred partly by reciprocal recombination at solo delta sequences. We identified a new open reading frame sufficient to code for a 554 amino acid long protein of unknown function. The new open reading frame (Accession number Z37509) is located in the 3′ non-coding region of MEL3-MEL10 genes in opposite orientation to the MEL genes (Accession numbers Z37508, Z37510, Z37511). Northern analysis of total RNA showed no hybridization to a homologous probe, suggesting that the gene is not expressed efficiently if at all.     

Rate of heat release and ignitability indices for surface linings

Dimensional analysis is applied to one form of fire growth in a compartment to derive two indices for the classification of linings, an ignitability index and a rate of heat release index, that can be combined into a fire growth parameter, which here is made proportional to the time to flashover in the internationally standardized Room/Corner Test. The ignitability index is the inverse of the time to ignition. The rate of heat release index is obtained by integrating the rate of heat release in time, weighted in such a way that the values of rate of heat release immediately after ignition are of higher importance than those at later times. Both indices are determined from test results of the Cone Calorimeter.