A Structured Prediction Approach for Statistical Machine Translation

We propose a new formally syntax-based method for statistical machine translation. Transductions between parsing trees are transformed into a problem of sequence tagging, which is then tackled by a search- based structured prediction method. This allows us to automatically acquire transla- tion knowledge from a parallel corpus without the need of complex linguistic parsing. This method can achieve compa- rable results with phrase-based method (like Pharaoh), however, only about ten percent number of translation table is used. Experiments show that the structured pre- diction approach for SMT is promising for its strong ability at combining words.

基于结构转换和序列标注的形式化句法统计翻译模型研究

随着全球化的不断深入和互联网的发展，人们在日常工作和生活中常常需要面对大量的非母语信息，如何利用计算机实现不同语言之间的自动转换以克服人类的语言障碍已成为当前人们的迫切需求。由于统计机器翻译方法具有语言依赖性弱、系统开发周期短、翻译质量较好等优点，目前已成为机器翻译研究界的热点研究方向。本文从形式化句法的角度，针对目前统计机器翻译方法中存在的非连续短语处理、训练和搜索相独立以及短语顺序的调整等问题开展了研究工作。论文的主要工作和创新点总结如下： 1. 改进的短语翻译模型 目前常用的基于短语的翻译模型没有考虑对非连续短语的处理。我们提出了一种基于非连续短语的统计翻译模型，利用该模型可以使翻译的基本单元从连续短语扩展到带有间隔的非连续短语，通过借助上下文词汇信息以达到改善翻译结果的目的。同时，由于该方法抽取的短语数量较少，也使得解码的效率得到了提高。实验表明，改进的非连续短语模型在取得与层次型短语模型相当的翻译性能的情况下，计算效率也得到了提高。 2. 基于SEARN的形式化句法模型 在目前的机器学习方法中，训练和搜索的过程相对独立，训练时采用的复杂结构信息在搜索过程中常常难于保持。我们提出了一种转换操作，将集成训练和搜索的结构化预测方法（Search-based Structured Prediction，SEARN）用于转换后的序列标注问题，以解决双语句法树之间的映射。实验表明，该方法在抽取短语数量只有短语模型十分之一的情形下，仍可取得与短语模型相当的翻译性能。 3. 基于压平的双语句法树结构的形式化句法模型 短语顺序是翻译时需要关注的关键问题之一，目前的方法通常只是以源语言端的信息作为调序的依据。我们提出了一种基于压平的双语句法树结构的形式化句法方法。其核心是抽取带有方向属性的短语对，利用这种方向属性辅助目标句子的生成，从而改善目标句子的内部结构（短语顺序）以提高翻译质量。在NIST MT08机器翻译评测数据上的实验表明，这种方法和基于短语的系统相比BLUE值获得7%的提高。 4. 基于序列标注的形式化句法模型 在翻译时句子中的某些区域通常倾向于作为一个整体来进行翻译，目前的方法对句子中的所有词都允许任意位置的词序调整，带来了很多不合理的顺序调整结果。我们提出了一种基于序列标注的形式化句法模型。首先利用压平的双语句法树结构表示双语句子，然后通过我们定义的标签对树节点进行标记，最后借助条件随机场模型对这些标签进行学习。通过这些标签可以区分出能够作为整体翻译的区域，以及句子中难于翻译的部分。同时，对不同的翻译区域可以采用不同的翻译方法，通过局部翻译质量的提高可以带来了整个句子翻译质量的改善。该模型和基于层次型短语的模型比较，BLUE值得到了5%的提高。

State Vector: A New Approach to Prediction of the Failure of Brittle Heterogeneous Media and Large Earthquakes

The concept of state vector stems from statistical physics, where it is usually used to describe activity patterns of a physical field in its manner of coarsegrain. In this paper, we propose an approach by which the state vector was applied to describe quantitatively the damage evolution of the brittle heterogeneous systems, and some interesting results are presented, i.e., prior to the macro-fracture of rock specimens and occurrence of a strong earthquake, evolutions of the four relevant scalars time series derived from the state vectors changed anomalously. As retrospective studies, some prominent large earthquakes occurred in the Chinese Mainland (e.g., the M 7.4 Haicheng earthquake on February 4, 1975, and the M 7.8 Tangshan earthquake on July 28, 1976, etc) were investigated. Results show considerable promise that the time-dependent state vectors could serve as a kind of precursor to predict earthquakes.

Perturbational Finite Volume Method For The Solution of 2-D Navier-Stokes Equations on Unstructured and Structured Colocated Meshes

Based on the first-order upwind and second-order central type of finite volume( UFV and CFV) scheme, upwind and central type of perturbation finite volume ( UPFV and CPFV) schemes of the Navier-Stokes equations were developed. In PFV method, the mass fluxes of across the cell faces of the control volume (CV) were expanded into power series of the grid spacing and the coefficients of the power series were determined by means of the conservation equation itself. The UPFV and CPFV scheme respectively uses the same nodes and expressions as those of the normal first-order upwind and second-order central scheme, which is apt to programming. The results of numerical experiments about the flow in a lid-driven cavity and the problem of transport of a scalar quantity in a known velocity field show that compared to the first-order UFV and second-order CFV schemes, upwind PFV scheme is higher accuracy and resolution, especially better robustness. The numerical computation to flow in a lid-driven cavity shows that the under-relaxation factor can be arbitrarily selected ranging from 0.3 to 0. 8 and convergence perform excellent with Reynolds number variation from 102 to 104.

Load-Unload Response Ratio and Accelerating Moment/Energy Release Critical Region Scaling and Earthquake Prediction

The main idea of the Load-Unload Response Ratio (LURR) is that when a system is stable, its response to loading corresponds to its response to unloading, whereas when the system is approaching an unstable state, the response to loading and unloading becomes quite different. High LURR values and observations of Accelerating Moment/Energy Release (AMR/AER) prior to large earthquakes have led different research groups to suggest intermediate-term earthquake prediction is possible and imply that the LURR and AMR/AER observations may have a similar physical origin. To study this possibility, we conducted a retrospective examination of several Australian and Chinese earthquakes with magnitudes ranging from 5.0 to 7.9, including Australia's deadly Newcastle earthquake and the devastating Tangshan earthquake. Both LURR values and best-fit power-law time-to-failure functions were computed using data within a range of distances from the epicenter. Like the best-fit power-law fits in AMR/AER, the LURR value was optimal using data within a certain epicentral distance implying a critical region for LURR. Furthermore, LURR critical region size scales with mainshock magnitude and is similar to the AMR/AER critical region size. These results suggest a common physical origin for both the AMR/AER and LURR observations. Further research may provide clues that yield an understanding of this mechanism and help lead to a solid foundation for intermediate-term earthquake prediction.

Prediction of structural dynamic plastic shear failure by Johnson's damage number

It is proved that Johnson's damage number is the sole similarity parameter for dynamic plastic shear failure of structures loaded impulsively, therefore, dynamic plastic shear failure can be understood when damage number reaches a critical value. It is suggested that the damage number be generally used to predict the dynamic plastic shear failure of structures under various kinds of dynamic loads (impulsive loading, rectangular pressure pulse, exponential pressure pulse, etc.,). One of the advantages for using the damage number to predict such kind of failure is that it is conveniently used for dissimilar material modeling.

High Aspect Ratio (L/D) Riser Viv Prediction Using Wake Oscillator Model

A two-dimensional (2-D) vortex-induced vibration (VIV) prediction model for high aspect ratio (LID) riser subjected to uniform and sheared flow is studied in this paper. The nonlinear structure equations are considered. The near wake dynamics describing the fluctuating nature of vortex shedding is modeled using classical van der Pol equation. A new approach was applied to calibrate the empirical parameters in the wake oscillator model. Compared the predicted results with the experimental data and computational fluid dynamic (CFD) results. Good agreements are observed. It can be concluded that the present model can be used as simple computational tool in predicting some aspects of VIV of long flexible structures. (C) 2008 Elsevier Ltd. All rights reserved.

Fast Computing For Lurr Of Earthquake Prediction

The LURR theory is a new approach for earthquake prediction, which achieves good results in earthquake prediction within the China mainland and regions in America, Japan and Australia. However, the expansion of the prediction region leads to the refinement of its longitude and latitude, and the increase of the time period. This requires increasingly more computations, and the volume of data reaches the order of GB, which will be very difficult for a single CPU. In this paper, a new method was introduced to solve this problem. Adopting the technology of domain decomposition and parallelizing using MPI, we developed a new parallel tempo-spatial scanning program.

On the influence zone and the prediction of tensile strength of particulate polymeric composites

An intended numerical investigation is carried out. The results indicate that, even if a perfect adhesive bond is preserved between the particles and matrix materials, the two-phase element cell model is unable to predict the strength increment of the particulate polymeric composites (PPC). To explore the main reinforcing mechanism, additional microscopic experiment is performed. An ''influence zone'' was observed around each particle which is measured about 2 to 10 micrometers in thickness for a glass-polyethylene mixture. Then, an improved computational model is presented to include the ''influence zone'' effect and several mechanical behaviors of PPC are well simulated through this new model.

Prediction of mixed-mode cracking direction in random, short-fibre composite-materials

The prediction of cracking direction in composite materials is of significance to the design of composite structures. This paper presents several methods for predicting the cracking direction in the double grooved tension-shear specimen which gives mixed-mode cracking. Five different criteria are used in this analysis: two of them have been used by other investigators and the others are proposed by the present authors. The strain energy density criterion proposed by G.C. Sih is modified to take account of the influence of the anisotropy of the strength on the direction of crack. The two failure criteria of Tsai-Hill and Norris are extended to predict the crack orientation. The stress distributions in the near-notch zone are calculated by using the 8-node quadrilateral isoparametric finite element method. The predictions of all the criteria except one are in good agreement with the experimental measurement. In addition, on the basis of the FEM results, the size of the zone in which the singular term is dominant is estimated.

Prediction of particle distribution in isotropic turbulence by large-eddy simulation

The recent application of large-eddy simulation (LES) to particle-laden turbulence requires that the LES with a subgrid scale (SGS) model could accurately predict particle distributions. Usually, a SGS particle model is used to recover the small-scale structures of velocity fields. In this study, we propose a rescaling technique to recover the effects of small-scale motions on the preferential concentration of inertial particles. The technique is used to simulate particle distribution in isotropic turbulence by LES and produce consistent results with direct numerical simulation (DNS). Key words: particle distribution, particle-laden turbulence, large-eddy simulation, subgrid scale model.

Prediction Of Shear-Band Thickness In Metallic Glasses

We derive an explicit expression for predicting the thicknesses of shear bands in metallic glasses. The model demonstrates that the shear-band thickness is mainly dominated by the activation size of the shear transformation zone (STZ) and its activation free volume concentration. The predicted thicknesses agree well with the results of measurements and simulations. The underlying physics is attributed to the local topological instability of the activated STZ. The result is of significance in understanding the origin of inhomogeneous flow in metallic glasses. (C) 2009 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved.

Prediction of Effective Moduli of Carbon Nanotube-Reinforced Composites with Waviness and Debonding

Carbon nanotubes (CNTs) have been regarded as ideal reinforcements of high-performance composites with enormous applications. However, the waviness of the CNTs and the interfacial bonding condition between them and the matrix are two key factors that influence the reinforcing efficiency. In this paper, the effects of the waviness of the CNTs and the interfacial debonding between them and the matrix on the effective moduli of CNT-reinforced composites are studied. A simple analytical model is presented to investigate the influence of the waviness on the effective moduli. Then, two methods are proposed to examine the influence of the debonding. It is shown that both the waviness and debonding can significantly reduce the stiffening effect of the CNTs. The effective moduli are very sensitive to the waviness when the latter is small, and this sensitivity decreases with the increase of the waviness. (C) 2008 Elsevier Ltd. All rights reserved.