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.     

Atomic layer deposition of calcium oxide and calcium hafnium oxide films using calcium cyclopentadienyl precursor

Calcium oxide and calcium hafnium oxide thin films were grown by atomic layer deposition on borosilicate glass and silicon substrates in the temperature range of 205–300 °C. The calcium oxide films were grown from novel calcium cyclopentadienyl precursor and water. Calcium oxide films possessed refractive index 1.75–1.80. Calcium oxide films grown without Al₂O₃ capping layer occurred hygroscopic and converted to Ca(OH)₂ after exposure to air. As-deposited CaO films were (200)-oriented. CaO covered with Al₂O₃ capping layers contained relatively low amounts of hydrogen and re-oriented into (111) direction upon annealing at 900 °C. In order to examine the application of CaO in high-permittivity dielectric layers, mixtures of Ca and Hf oxides were grown by alternate CaO and HfO₂ growth cycles at 230 and 300 °C. HfCl₄ was used as a hafnium precursor. When grown at 230 °C, the films were amorphous with equal amounts of Ca and Hf constituents (15 at.%). These films crystallized upon annealing at 750 °C, showing X-ray diffraction peaks characteristic of hafnium-rich phases such as Ca₂Hf₇O₁₆ or Ca₆Hf₁₉O₄₄. At 300 °C, the relative Ca content remained below 8 at.%. The crystallized phase well matched with rhombohedral Ca₂Hf₇O₁₆. The dielectric films grown on Si(100) substrates possessed effective permittivity values in the range of 12.8–14.2.     

Identifying software project risks in Nigeria: an International Comparative Study

We report a study where software project risk factors were solicited in Nigeria. The study repeats a research design, which has been used in a Delphi study in the U.S., Finland, and Hong Kong to rank software development risks. The study produces a rank-order list of software risk factors and compares results with the earlier study, showing significant differences. Obtained rankings signal the importance of the infrastructure related and socio-economic software risks in developing countries like Nigeria. This demonstrates the importance of understanding the broader socio-economic context in identifying and managing software risks. The study also shows that some risks in Nigeria are located on a different scale. The choice of contingent mitigation strategies of project managers play a less important role, because the risks are so deeply engrained with the poor quality of basic infrastructure services.     

Weight initialization with reference patterns

The problem of weight initialization in multilayer perceptron networks is considered. A new computationally simple weight initialization method based on the usage of reference patterns is presented. A reference pattern is a vector which is used to represent data points that fall in its vicinity in the data space. On one hand, the proposed method aims to set the initial weight values to be such that inputs to network nodes are within the active region (in other words, nodes are not saturated). On the other hand, the goal is to distribute the discriminant functions formed by the hidden units evenly into the input space area where training data is located. The proposed method is tested with the widely used two-spirals classification benchmark problem and channel equalization problem where several alternatives for obtaining suitable reference patterns are investigated. Also, the effect of the initialization is studied when two commonly used cost functions are used in the training phase. These are the mean square error and relative entropy cost functions. A comparison with the conventional random initialization shows that significant improvement in convergence can be achieved with the proposed method. In addition, the computational cost of the initialization was found to be negligible compared with the cost of training.     

化学分离法分析大型纸浆发电厂废渣中生物质底灰和飞灰成分(英文)

采用三步分离法,其中第一步用水,第二步用乙酸铵,第三步用盐酸,提取纸浆发电厂生物质飞灰和底灰的组分,以评估这些残渣的浸出行为。除了Mo、S、Na以及那些浓度低于检测范围的元素外,在这两种灰份中,对于其他元素这三种浸出剂的浸出能力从弱到强的顺序为:水、乙酸铵和盐酸。采用三步分离法处理的废渣中元素浓度比前期用BCR连续提取法得到的低。     

Environmental Control of Triplet Emission in Donor–Bridge–Acceptor Organometallics

Carbene‐metal‐amides (CMAs) are a promising family of donor–bridge–acceptor molecular charge‐transfer (CT) emitters for organic light‐emitting diodes. A universal approach is demonstrated to tune the energy of their CT emission. A blueshift of up to 210 meV is achievable in solid state via dilution in a polar host matrix. The origin of this shift has two components: constraint of thermally‐activated triplet diffusion, and electrostatic interactions between guest and polar host. This allows the emission of mid‐green CMA archetypes to be tuned to sky blue without chemical modifications. Monte‐Carlo simulations based on a Marcus‐type transfer integral successfully reproduce the concentration‐ and temperature‐dependent triplet diffusion process, revealing a substantial shift in the ensemble density of states in polar hosts. In gold‐bridged CMAs, this shift does not lead to a significant change in luminescence lifetime, thermal activation energy, reorganization energy, or intersystem crossing rate. These discoveries offer new insight into coupling between the singlet and triplet manifolds in CMA materials, revealing a dominant interaction between states of CT character. The same approach is employed using materials which have been chemically modified to alter the energy of their CT state directly, shifting the emission of sky‐blue chromophores into the practical blue range.     

Strategies for software-based hybrid business models

The open source approach to software development has been used by software organizations in tandem with their existing business models, which are based on proprietary software licensing. This led to the creation of hybrid business models that merge open source and proprietary paradigms. This paper explores the practices used by software product vendors using hybrid business models and proposes strategies emerging out of these practices using interpretive, single case study research design.