新闻领域双语语料建设与句子对齐方法的研究

双语对齐的平行语料库在机器翻译和自然语言处理领域中扮演着非常重要的角色,它的研究和制作具有重要的理论意义和实用价值,双语语料的建设十分必要,其中双语对齐是最基本的环节.文章首先简要介绍了语料库的建设情况,然后结合主流的句子对齐方法提出并实现了基于词典和语言学信息的英汉双语句子对齐.     

基于MDL Shape Model及EFD的行人

针对智能监控的行人轮廓特征提取,先利用MDL shape model得到紧致但能充分表征轮廓几何特征标志点之间的对应关系,再利用椭圆傅里叶分解把行人轮廓表示为一系列不同频率下椭圆傅里叶系数组成的向量,最后得到行人2D＋time轮廓的描述向量。实验结果表明,此方法不但可以在几何上直观地表示行人轮廓,而且大大降低了轮廓向量的维数。     

传递对准中时间延迟的补偿方法

在机载战术导弹动基座传递对准过程中,主惯导数据传递给弹载计算机有时间延迟因而会影响传递对准中卡尔曼滤波器的收敛速度和精度.研究分析了从机载火控计算机所传递的主惯导数据以及这些数据的延迟时间,提出直接利用主惯导姿态四元素、载机角速度和延迟时间等主惯导数据进行四元素计算,得出延迟后的主惯导姿态四元素,进而获得补偿后的量测失准角的算法,并建立基于"速度 姿态"匹配方式的传递对准仿真模型进行验证,仿真结果表明该算法能有效的抑制时间延迟所造成的影响.     

强跟踪滤波器在潜射运载器传递对准中的应用

潜射运载器在发射前进行精确的传递对准十分重要.然而潜射运载器的传递对准精度会因为潜艇挠曲变形而受到严重影响.为了解决艇体挠曲变形给运载器传递对准精度带来的不利影响,对强跟踪滤波器在潜射运载器动基座传递对准中的应用进行了研究,设计了强跟踪滤波器,仿真结果表明在潜艇具有很小的机动以及考虑艇体挠曲变形的条件下能够较好地估计出从惯导相对主惯导的失准角.由于该方法不必建立艇体挠曲变形的复杂模型,使得解决挠曲变形对传递对准精度影响的问题大大简化了,具有一定的工程意义.     

Linear-space algorithms that build local alignments from fragments

This paper presents practical algorithms for building an alignment of two long sequences from a collection of alignment fragments, such as all occurrences of identical 5-tuples in each of two DNA sequences. We first combine a time-efficient algorithm developed by Galil and coworkers with a space-saving approach of Hirschberg to obtain a local alignment algorithm that uses0((M+N+F logN) logM) time and0(M+N) space to align sequences of lengthsM andN from a pool of F alignment fragments. Ideas of Huang and Miller are then employed to develop a time- and space-efficient algorithm that computesn best nonintersecting alignments for anyn>1. An example illustrates the utility of these methods.This work was supported in part by Grant RO1 LM05110 from the National Library of Medicine.     

融合词簇约束的汉越跨语言词嵌入

针对传统跨语言词嵌入方法在汉越等差异较大的低资源语言上对齐效果不佳的问题,提出一种融合词簇对齐约束的汉越跨语言词嵌入方法。通过独立的单语语料训练获取汉越单语词嵌入,使用近义词、同类词和同主题词3种不同类型的关联关系,充分挖掘双语词典中的词簇对齐信息以融入到映射矩阵的训练过程中,使映射矩阵进一步学习到不同语言相近词间具有的一些共性特征及映射关系,根据跨语言映射将两种语言的单语词嵌入映射至同一共享空间中对齐,令具有相同含义的汉语与越南语词嵌入在空间中彼此接近,并利用余弦相似度为空间中每一个未经标注的汉语单词查找对应的越南语翻译构建汉越对齐词对,实现跨语言词嵌入。实验结果表明,与传统有监督及无监督的跨语言词嵌入方法Multi_w2v、Orthogonal、VecMap、Muse相比,该方法能有效提升映射矩阵在非标注词上的泛化性,改善汉越低资源场景下模型对齐效果较差的问题,其在汉越双语词典归纳任务P@1和P@5上的对齐准确率相比最好基线模型提升了2.2个百分点。     

不完全匹配的语音和文本语句级对齐

语音文本自动对齐技术广泛应用于语音识别与合成、内容制作等领域,其主要目的是将语音和相应的参考文本在语句、单词、音素等级别的单元进行对齐,并获得语音与参考文本之间的时间对位信息.最新的先进对齐方法大多基于语音识别,一方面,准确率受限于语音识别效果,识别字错误率高时文语对齐精度明显下降,识别字错误率对对齐精度影响较大;另一方面,这种对齐方法不能有效处理不完全匹配的长篇幅语音和文本的对齐.该文提出一种基于锚点和韵律信息的文语对齐方法,通过基于边界锚点加权的片段标注将语料划分为对齐段和未对齐段,针对未对齐段使用双门限端点检测方法提取韵律信息,并检测语句边界,降低了基于语音识别的对齐方法对语音识别效果的依赖程度.实验结果表明,与目前先进的基于语音识别的文语对齐方法比较,即使在识别字错误率为0.52时,该文所提方法的对齐准确率仍能提升45%以上;在音频文本不匹配程度为0.5时,该文所提方法能提高3%.     

事件日志的批量迹与过程模型的多视角对齐方法

在业务过程发现的一致性检测中,现有事件日志与过程模型的多视角对齐方法一次只能获得一条迹与过程模型的最优对齐;并且最优对齐求解中的启发函数计算复杂,以致最优对齐的计算效率较低。为此,提出一种基于迹最小编辑距离的、事件日志的批量迹与过程模型的多视角对齐方法。首先选取事件日志中的多条迹组成批量迹,使用过程挖掘算法得到批量迹的日志模型;进而获取日志模型与过程模型的乘积模型及其变迁系统,即为批量迹的搜索空间;然后设计基于Petri网变迁序列集合与剩余迹的最小编辑距离的启发函数来加快A*算法;最后设计可调节数据和资源视角所占权重的多视角代价函数,在乘积模型的变迁系统上提出批量迹中每条迹与过程模型的多视角最优对齐方法。仿真实验结果表明,相比已有工作,在计算批量迹与过程模型间的多视角对齐时,所提方法占用更少的内存空间和使用更少的运行时间。该方法提高了最优对齐的启发函数计算速度,可以一次获得批量迹的所有最优对齐,进而提高了事件日志与过程模型的多视角对齐效率。     

Precisely Aligned Monolayer MoS2 Epitaxially Grown on h‐BN basal Plane

Control of the precise lattice alignment of monolayer molybdenum disulfide (MoS₂) on hexagonal boron nitride (h‐BN) is important for both fundamental and applied studies of this heterostructure but remains elusive. The growth of precisely aligned MoS₂ domains on the basal plane of h‐BN by a low‐pressure chemical vapor deposition technique is reported. Only relative rotation angles of 0° or 60° between MoS₂ and h‐BN basal plane are present. Domains with same orientation stitch and form single‐crystal, domains with different orientations stitch and from mirror grain boundaries. In this way, the grain boundary is minimized and a continuous film stitched by these two types of domains with only mirror grain boundaries is obtained. This growth strategy is also applicable to other 2D materials growth.     

Mechanical and electrical properties of electrospun CNT/PVDF nanofiber for micro-actuator applications

Electrospun polyvinylidene fluoride (PVDF)-containing carbon nanotubes (CNT) were prepared for use in fabricating actuator materials. Actuating displacement was measured in an electrochemical environment. The electrospun nanofibers were arranged using a drum-type collector, and morphology was investigated using a field emission-scanning electron microscope. The uniformity of dispersion of CNT in the PVDF nanofibers was monitored by electron probe X-ray micro-analysis. Tensile strength and electrical resistivity results were used as an indication of the state of alignment. The electrospun CNT/PVDF nanofiber sheets exhibited better mechanical and electrical properties in the arranged direction. The efficiency and electrical capacities of electrospun CNT/PVDF nanofiber sheet were compared with those of cast PVDF sheets for use in actuator applications in electrochemical environments. The electrospun CNT/PVDF nanofiber sheets exhibited much better actuator performance than PVDF sheets, which are attributed to their superior electrical properties.

Highlights

(1) The interfacial durability of CNT/PVDF nanofibers was enhanced to increase contact area by reinforcing CNT.

(2) The efficiency of CNT/PVDF actuators was improved due to interfacial properties.

(3) Thin thickness drum-type collector was made to enhance nanofiber alignment.

(4) The arranged CNT/PVDF nanofibers exhibited better mechanical and actuating displacements.