神经机器译前沿进展.pptVIP

神经机器翻译前沿进展  机器翻译 ·目标:利用计算机实现自然语言的自动翻译 布什与沙龙举行了「会谈 只 Bush held talk WI Sharon  发展历史 趋势:让机器更“自主”地学习如何翻译 规则 数据驱动 机器翻译 机器翻译 198 1990  数据驱动的机器翻译 ·核心问题:如何为翻译过程建立概率模型? x布什与沙龙」(举行『了」会谈 几 y Bush held a talkwith Sharon  统计机器翻译 隐变量对数线性模型:在隐式语言结构上设计特征 x布什与沙龙」(举行『了」会谈 p=x∑0e,ky,2 (Och and Ney, 2002) y Bush held a talkwith Sharon  基于短语的统计机器翻译 ·短语翻译模型:以隐结构短语为基本翻译单元 布什与沙龙举行 会谈 布什 与沙龙 举行了会谈 Bush ith sharon held a talk Bush held a talk with Sharon Sharo Bush held talk with n (Koehn et al, 2003)  统计机器翻译的优缺点 优点 隐结构可解释性高 利用局部特征和动态规划处理指数级结构空间 缺点 线性模型难以处理高维空间中线性不可分的情况 需要人类专家设计隐式结构及相应的翻译过程 需要人类专家设计特征 离散表示带来严重的数据稀疏问题 难以处理长距离依赖  难点:长距离调序 held Bus President talk Sharon Ministe Israel the 1+ Prim House White 如何用上述词语拼成合理的译文?  统计机器翻译示例 Chinese 美国总统布什昨天在白宫与以色列总理沙龙就中东局势X 行了一个小时 会谈 English Yesterday, U.S. President George W. Bush at the White situation in the Middle East held a one-hour talk House with Israeli Prime Minister Ariel Sharon on  深度学习带来新思路 nature full stop is chosen//. Overall, this process generates sequences of rench words according to a probability distribution that depends on the English sentence. This rather naive way of performing machine translation has quickly become competitive with the state-of-the-art, and this raises serious doubts about whether understanding a sen- tence requires anything like the internal symbolic expressions that are manipulated by using inference rules. It is more compatible with the Yann lecun Yoshua bengio Geoffrey hinton (LeCun et al, 2015)