融合单词翻译的神经机器翻译

神经机器翻译由于无法完全学习源端单词语义信息,往往造成翻译结果中存在着大量的单词翻译错误。该文提出了一种融入单词翻译用以增强源端信息的神经机器翻译方法。首先使用字典方法找到每个源端单词对应的目标端翻译,然后提出并比较两种不同的方式,用以融合源端单词及其翻译信息: ①Factored 编码器: 单词及其翻译信息直接相加; ②Gated 编码器: 通过门机制控制单词翻译信息的输入。基于目前性能最优的基于自注意力机制的神经机器翻译框架Transformer,在中英翻译任务的实验结果表明,与基准系统相比,该文提出的两种融合源端单词译文的方式均能显著提高翻译性能,BLEU值获得了0.81个点的提升。     

Automatic Extraction of Rules for Anaphora Resolution of Japanese Zero Pronouns in Japanese–English Machine Translation from Aligned Sentence Pairs

This paper proposes a method to extract rules for the anaphora resolution of Japanese zero pronouns in Japanese–English MT from aligned sentence pairs. After aligned sentence pairs unsuitable for rule extraction because of analysis errors or free translations are automatically rejected, zero pronouns in the Japanese sentences and the English translation equivalents of their antecedents are extracted from the remaining Japanese and English aligned sentence pairs using ten hand-developed alignment rules. This method identifies all Japanese zero pronouns whose translation equivalents are not explicitly expressed in an English sentence, this method identifies these as unalignable. Then, resolution rules for the remaining zero pronouns are automatically extracted using the aligned pairs, equivalent word/phrase pairs extracted from the aligned sentence pairs, and the syntactic and semantic structures of the Japanese sentences. This method was implemented in a Japanese–English MT system, ALT-J/E. 98.4% of all pairs were automatically aligned correctly in a window test, and 94.0% in a blind test. Furthermore, extracted rules for zero pronouns with deictic references created automatically from sentence pairs correctly resolved 99.0% of the zero pronouns in a window test and 85.0% of the zero pronouns in a blind test.     

A Technical Word- and Term-Translation Aid Using Noisy Parallel Corpora across Language Groups

Technical-term translation represents one of the most difficult tasks for human translators since (1) most translators are not familiar with terms and domain-specific terminology and (2) such terms are not adequately covered by printed dictionaries. This paper describes an algorithm for translating technical words and terms from noisy parallel corpora across language groups. Given any word which is part of a technical term in the source language, the algorithm produces a ranked candidate match for it in the target language. Potential translations for the term are compiled from the matched words and are also ranked. We show how this ranked list helps translators in technical-term translation. Most algorithms for lexical and term translation focus on Indo-European language pairs, and most use a sentence-aligned clean parallel corpus without insertion, deletion or OCR noise. Our algorithm is language- and character-set-independent, and is robust to noise in the corpus. We show how our algorithm requires minimum preprocessing and is able to obtain technical-word translations without sentence-boundary identification or sentence alignment, from the English–Japanese awk manual corpus with noise arising from text insertions or deletions and on the English–Chinese HKUST bilingual corpus. We obtain a precision of 55.35% from the awk corpus for word translation including rare words, counting only the best candidate and direct translations. Translation precision of the best-candidate translation is 89.93% from the HKUST corpus. Potential term translations produced by the program help bilingual speakers to get a 47% improvement in translating technical terms.     

Large-scale automatic extraction of an English-Chinese translation lexicon

We report experimental results on automatic extraction of an English-Chinese translation lexicon, by statistical analysis of a large parallel corpus, using limited amounts of linguistic knowledge. To our knowledge, these are the first empirical results of the kind between an Indo-European and non-Indo-European language for any significant vocabulary and corpus size. The learned vocabulary size is about 6,500 English words, achieving translation precision in the 86–96% range, with alignment proceeding at paragraph, sentence, and word levels. Specifically, we report (1) progress on the HKUST English-Chinese Parallel Bilingual Corpus, (2) experiments supporting the usefulness of restricted lexical cues for statistical paragraph and sentence alignment, and (3) experiments that question the role of hand-derived monolingual lexicons for automatic word translation acquisition. Using a hand-derived monolingual lexicon, the learned translation lexicon averages 2.33 Chinese translations per English entry, with a manually-filtered precision of 95.1%, and an automatically-filtered weighted precision of 86.0%. We then introduce a fully automatic two-stage statistical methodology that is able to learn translations for collocations. A statistically-learned monolingual Chinese lexicon is first used to segment the Chinese text, before applying bilingual training to produce 6,429 English entries with 2.25 Chinese translations per entry. This method improves the manually-filtered precision to 96.0% and the automatically-filtered weighted precision to 91.0%, an error rate reduction of 35.7% from using a hand-derived monolingual lexicon.     

基于词对关联网络的句子对齐研究

句子对齐能够为跨语言的自然语言处理任务提供高质量的对齐句子对。受对齐句子对通常包含大量对齐的单词对这种直觉的启发,该文通过探索神经网络框架下词对间的语义相互作用来解决句子对齐问题。特别地,该文提出的词对关联网络通过融合三种相似性度量方法从不同角度来捕获词对之间的语义关系,并进一步融合它们之间的语义关系来确定两个句子是否对齐。在单调和非单调文本上的实验结果表明,该文提出的方法显著提高了句子对齐的性能。     

基于语料库的最小求交词对齐

提出了一种基于句子级对齐的双语语料库的英汉词对齐方法。它建立在句对的集合表示形式的基础上。通过最小求交模型实现词对齐。使用倒排索引表和集合运算实现高效的最小求交算法。在对齐过程中引入高频干扰词表以提高召回率。实验结果表明，该方法优于使用共现互信息的词对齐和使用双语词典的词对齐方法。     

Complexity of alignment and decoding problems: restrictions and approximations

We study the computational complexity of the Viterbi alignment and relaxed decoding problems for IBM model 3, focusing on the problem of finding a solution which has significant overlap with an optimal. That is, an approximate solution is considered good if it looks like some optimal solution with a few mistakes, where mistakes can be wrong values (such as a word aligned incorrectly or a wrong word in decoding), as well as insertions and deletions (spurious/missing words in decoding). In this setting, we show that it is computationally hard to find a solution which is correct on more than half (plus an inverse polynomial fraction) of the words. More precisely, if there is a polynomial-time algorithm computing an alignment for IBM model 3 which agrees with some Viterbi alignment on \(l/2+l^\epsilon \) words, where l is the length of the English sentence, or producing a decoding with \(l/2+l^\epsilon \) correct words, then P \(=\) NP. We also present a similar structure inapproximability result for phrase-based alignment. As these strong lower bounds are for the general definitions of the Viterbi alignment and decoding problems, we also consider, from a parameterized complexity perspective, which properties of the input make these problems intractable. As a first step in this direction, we show that Viterbi alignment has a fixed-parameter tractable algorithm with respect to limiting the range of words in the target sentence to which a source word can be aligned. We note that by comparison, limiting maximal fertility—even to three—does not affect NP-hardness of the result.     

句对齐研究综述

神经机器翻译是目前机器翻译领域的主流方法,拥有足够数量的双语平行语料是训练出一个好的翻译模型的前提。双语句对齐技术作为一种从不同语言端单语语料中获取双语平行句对的技术,因此得到广泛的研究。该文首先简单介绍句对齐任务及其相应的评测标准,然后归纳总结前人在句对齐任务上的研究进展,以及句对齐任务的相关信息,并简单概括参加团队所提交的系统,最后对当前工作进行总结并展望未来的工作。     

TransSearch: from a bilingual concordancer to a translation finder

As basic as bilingual concordancers may appear, they are some of the most widely used computer-assisted translation tools among professional translators. Nevertheless, they still do not benefit from recent breakthroughs in machine translation. This paper describes the improvement of the commercial bilingual concordancer TransSearch in order to embed a word alignment feature. The use of statistical word alignment methods allows the system to spot user query translations, and thus the tool is transformed into a translation search engine. We describe several translation identification and postprocessing algorithms that enhance the application. The excellent results obtained using a large translation memory consisting of 8.3 million sentence pairs are confirmed via human evaluation.     

融入分类词典的汉越混合网络神经机器翻译集外词处理方法

在神经机器翻译中,因词表受限导致的集外词问题很大程度上影响了翻译系统的准确性。对于训练语料较少的资源稀缺型语言的神经机器翻译,这种问题表现得更为严重。近几年,受到外部知识融入的启发,该文在RNNSearch模型基础上,提出了一种融入分类词典的汉越混合网络神经机器翻译集外词处理方法。对于给定的源语言句子,扫描分类词典以确定候选短语句对并标签标记,解码端利用词级组件和短语组件的混合解码网络,很好地生成单词集外词和短语集外词的翻译,从而改善汉越神经机器翻译的性能。在汉越、英越和蒙汉翻译实验上表明,该方法显著提高了准确率,对于资源稀缺型语言的神经机器翻译性能有一定的提升。     

System Combination for Machine Translation of Spoken and Written Language

This paper describes an approach for computing a consensus translation from the outputs of multiple machine translation (MT) systems. The consensus translation is computed by weighted majority voting on a confusion network, similarly to the well-established ROVER approach of Fiscus for combining speech recognition hypotheses. To create the confusion network, pairwise word alignments of the original MT hypotheses are learned using an enhanced statistical alignment algorithm that explicitly models word reordering. The context of a whole corpus of automatic translations rather than a single sentence is taken into account in order to achieve high alignment quality. The confusion network is rescored with a special language model, and the consensus translation is extracted as the best path. The proposed system combination approach was evaluated in the framework of the TC-STAR speech translation project. Up to six state-of-the-art statistical phrase-based translation systems from different project partners were combined in the experiments. Significant improvements in translation quality from Spanish to English and from English to Spanish in comparison with the best of the individual MT systems were achieved under official evaluation conditions.      

基于双语语料的单个源语词汇和目标语多词单元的对齐

多词单元包括固定搭配、多词习语和多词术语等。本文提供了一个基于双语口语语料库的自动对齐单个源语词汇和目标语多词单元的算法,算法一方面通过计算对应于同一个源语词汇,多个目标语词汇之间的互信息和t值的归一化差值的大小来衡量目标语多个词语之间的关联程度以提取多词单元,另一方面通过计算互信息和t值的平均值作为多词单元和单个源语词汇之间互为相互翻译的衡量程度,用局部最优、首尾禁用词过滤以及长词优先等策略很好地解决了这个问题。另外,对短语翻译词典的分级,有效地减少了高级别词典中非正确翻译项的数目,使得翻译词典具有更好的实用性。     

基于统计的锚点词候选集的获取

段对齐是在双语语料库中把各个段和它们的相应译文建立起对应关系,进而为以后的句子级对齐、短语级对齐和词汇级对齐提供资源。它在整个对齐研究中起到一个承上启下的作用。利用锚点词信息完成段对齐是一种常用且有效的方法。锚点词要求数量少准确率高,同时更为重要的是它要求具有可以说明两个段之间有对应关系的相对明显的特征,这就是说并不是文章中的每个词都可以做锚点词,锚点词也不是越多越好,这也决定了获取过程中不能借助字典等辅助信息,而必须使用其它方法。文章提出一种新的锚点词候选集的获取方法,通过统计和相似计算来得到锚点词的候选集信息。通过控制统计串的出现频数和相似度的大小就可得到令人满意的可用的结果。实验结果表明,当取高阈值且高相似度时,就可得到很高的准确率。因而该方法是一种获取锚点词的有效方法。     

基于多特征融合和图匹配的维汉句子对齐

维吾尔语新闻网页与对应的中文翻译网页在内容上往往并非完全可比,主要表现为双语句子序列的错位甚至部分句子缺失,这给维汉句子对齐造成了困难。此外,作为新闻要素的人名地名很多是未登录词,这进一步增加了维汉句子对齐的难度。为了提高维汉词汇的匹配概率,作者自动提取中文人名、地名并翻译为维吾尔译名,构造双语名称映射表并加入维汉双语词典。然后用维文句中词典词对应的中文译词在中文句中进行串匹配,以避免中文分词错误,累计所有匹配词对得到双语句对的词汇互译率。最后融合数字、标点、长度特征计算双语句对的相似度。在所有双语句子相似度构成的矩阵上,使用图匹配算法寻找维汉平行句对,在900个句对上最高达到95.67%的维汉对齐准确率。     

基于判别式模型的蒙英词对齐方法

词对齐是自然语言处理领域的基础性研究课题之一。文中提出基于链式条件随机场(CRF)判别式模型的蒙古文-英文词对齐方法。该方法根据蒙古文和英文之间的差异,选择词形、词汇、词性等信息作为特征,建立双层CRF词对齐模型:在第1层利用CRF模型实现子块的对齐;在第2层利用CRF模型得到块内词之间的对齐结果。通过人工构建的词对齐语料开展相应实验。实验结果表明,该方法有效提高蒙英词对齐质量。     

基于多编码器多解码器的大规模维汉神经网络机器翻译模型

为提升维汉机器翻译模型的翻译能力,该文提出使用多编码器多解码器的结构,搭建大规模的维汉神经网络机器翻译模型。相比于单编码器单解码器的浅层的小模型,多编码器多解码器模型具有多个编码器,可以对源语言进行多层次、多视角的压缩表示;同时具有多个解码器,可以增强目标语言的生成能力。实验证明,在大规模的训练数据上,使用该方法搭建的大规模维汉神经网络机器翻译模型,译文质量可以大幅度地超过基于短语的统计机器翻译模型和基本的神经网络翻译模型。该文还针对维汉翻译源端语言和目标端语言的翻译单元粒度进行了实验,发现维吾尔语端使用字节对编码单元、汉语端使用字单元,可以消除对汉语分词器的依赖,做到和双端都使用字节对编码单元可比的效果。     

A Substitution-Translation-Restoration Framework for Handling Unknown Words in Statistical Machine Translation

Unknown words are one of the key factors that greatly affect the translation quality.Traditionally, nearly all the related researches focus on obtaining the translation of the unknown words.However, these approaches have two disadvantages.On the one hand, they usually rely on many additional resources such as bilingual web data;on the other hand, they cannot guarantee good reordering and lexical selection of surrounding words.This paper gives a new perspective on handling unknown words in statistical machine translation (SMT).Instead of making great efforts to find the translation of unknown words, we focus on determining the semantic function of the unknown word in the test sentence and keeping the semantic function unchanged in the translation process.In this way, unknown words can help the phrase reordering and lexical selection of their surrounding words even though they still remain untranslated.In order to determine the semantic function of an unknown word, we employ the distributional semantic model and the bidirectional language model.Extensive experiments on both phrase-based and linguistically syntax-based SMT models in Chinese-to-English translation show that our method can substantially improve the translation quality.     

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。     

基于动态词对齐的交互式机器翻译

在传统的机器翻译(machine translation,MT)与计算机辅助翻译(computer aided translation,CAT)中,译员与翻译引擎之间的交互受到很大限制,于是产生了交互式机器翻译(interactive machine translation,IMT)技术。但传统的模型只考虑当前源语与部分翻译的目标语的信息,没有将用户交互后的对齐信息加入到之后的预测模型中。该文基于词预测交互式机器翻译的研究思路,将用户交互翻译过程中的鼠标点选行为转化为中间译文的词对齐信息,进而在翻译交互过程中实现了对译文的动态词对齐标注,并在词对齐信息和输入译文的约束下提高了传统词预测的准确性。