基于多种知识的盲文翻译的研究

盲文到汉字的转换研究 ,分为盲文到拼音和拼音到汉字的转换研究。论文通过引入音码表 ,很好地解决了盲文到拼音转换中的歧义问题。对于拼音到汉字的转换 ,采用Markov模型 ,使用平滑算法解决此模型中的数据稀疏问题 ,并且对通常使用的字级别二元模型的平滑算法进行了改进以适应词级别二元模型。为每一个拼音句子建立一个拼音到汉字转换多部图 ,并在此转换多部图中使用 Viterbi算法搜索一条具有最大似然度的汉语句子 ,或使用 N - Best算法搜索具有头 N大似然度的 N个汉语句子。同时研究了从 1选到 N选的转换判据 ,以进行自动的 N - Best转换。实验表明 ,盲文到汉字转换系统的转换正确率为 94.38% ,若剔除专有名词 ,转换正确率还可上升 2个百分点。对 N - Best搜索算法的测试表明 5选正确率比 1选正确率提高了 3个百分点

Braille translation based on multi-knowledge

The transformation from Mandarin Braille to Mandarin Characters can be divided into two steps: from Braille to Mandarin Pinyin and from Pinyin to Mandarin Characters. Incorporating the Legal Pinyin Table into our system the ambiguity problem was solved in the transformation from Braille to Pinyin. A standard statistical bigram Markov model was used in the subsystem to transform Pinyin to Mandarin Characters. Then two modifications of the smoothing method which are consistent with the phrase level bigram model were proposed to overcome the sparse data problem in our system model.For each Pinyin sentence, a multi level graph was used with the Viterbi algorithm to search for the best Mandarin sentence in the maximal likelihood. The measurement of N best algorithm was studied to get N best Mandarin sentences. Experiments show that the correct rate of the system is 94.38%. If proper nouns were not considered, our system can achieve a further 2% improvement. The accuracy rate for the top 5 hypothesis by using N Best algorithm was 3% higher than that of the best hypothesis.