MIT LAB MTG MIT LAB MIT LAB MTG
![](https://slidetodoc.com/presentation_image_h2/d2072b75d2f1f05c564ad5a88929c16f/image-1.jpg)
![](http://slidetodoc.com/presentation_image_h2/d2072b75d2f1f05c564ad5a88929c16f/image-2.jpg)
![](http://slidetodoc.com/presentation_image_h2/d2072b75d2f1f05c564ad5a88929c16f/image-3.jpg)
![MI&T LAB MTG推导举例 MI&T LAB MTG推导举例](http://slidetodoc.com/presentation_image_h2/d2072b75d2f1f05c564ad5a88929c16f/image-4.jpg)
MI&T LAB MTG推导举例
![](http://slidetodoc.com/presentation_image_h2/d2072b75d2f1f05c564ad5a88929c16f/image-5.jpg)
![](http://slidetodoc.com/presentation_image_h2/d2072b75d2f1f05c564ad5a88929c16f/image-6.jpg)
![](http://slidetodoc.com/presentation_image_h2/d2072b75d2f1f05c564ad5a88929c16f/image-7.jpg)
![](http://slidetodoc.com/presentation_image_h2/d2072b75d2f1f05c564ad5a88929c16f/image-8.jpg)
![](http://slidetodoc.com/presentation_image_h2/d2072b75d2f1f05c564ad5a88929c16f/image-9.jpg)
![](http://slidetodoc.com/presentation_image_h2/d2072b75d2f1f05c564ad5a88929c16f/image-10.jpg)
![](http://slidetodoc.com/presentation_image_h2/d2072b75d2f1f05c564ad5a88929c16f/image-11.jpg)
![](http://slidetodoc.com/presentation_image_h2/d2072b75d2f1f05c564ad5a88929c16f/image-12.jpg)
![MI&T LAB 子图对 MI&T LAB 子图对](http://slidetodoc.com/presentation_image_h2/d2072b75d2f1f05c564ad5a88929c16f/image-13.jpg)
MI&T LAB 子图对
![](http://slidetodoc.com/presentation_image_h2/d2072b75d2f1f05c564ad5a88929c16f/image-14.jpg)
![](http://slidetodoc.com/presentation_image_h2/d2072b75d2f1f05c564ad5a88929c16f/image-15.jpg)
![](http://slidetodoc.com/presentation_image_h2/d2072b75d2f1f05c564ad5a88929c16f/image-16.jpg)
![MI&T LAB 子图MTG系统框架 input source language sentence Bilingual parse tress Language model Training system MI&T LAB 子图MTG系统框架 input source language sentence Bilingual parse tress Language model Training system](http://slidetodoc.com/presentation_image_h2/d2072b75d2f1f05c564ad5a88929c16f/image-17.jpg)
MI&T LAB 子图MTG系统框架 input source language sentence Bilingual parse tress Language model Training system Sub-graph pair table Decoding system output: bilingual “parse tree”
![](http://slidetodoc.com/presentation_image_h2/d2072b75d2f1f05c564ad5a88929c16f/image-18.jpg)
![](http://slidetodoc.com/presentation_image_h2/d2072b75d2f1f05c564ad5a88929c16f/image-19.jpg)
![](http://slidetodoc.com/presentation_image_h2/d2072b75d2f1f05c564ad5a88929c16f/image-20.jpg)
![](http://slidetodoc.com/presentation_image_h2/d2072b75d2f1f05c564ad5a88929c16f/image-21.jpg)
![](http://slidetodoc.com/presentation_image_h2/d2072b75d2f1f05c564ad5a88929c16f/image-22.jpg)
![](http://slidetodoc.com/presentation_image_h2/d2072b75d2f1f05c564ad5a88929c16f/image-23.jpg)
![](http://slidetodoc.com/presentation_image_h2/d2072b75d2f1f05c564ad5a88929c16f/image-24.jpg)
![MI&T LAB 解码算法 1 Vector<rule> XRules; 2 for l=1 to Sent. Length 3 for MI&T LAB 解码算法 1 Vector<rule> XRules; 2 for l=1 to Sent. Length 3 for](http://slidetodoc.com/presentation_image_h2/d2072b75d2f1f05c564ad5a88929c16f/image-25.jpg)
MI&T LAB 解码算法 1 Vector<rule> XRules; 2 for l=1 to Sent. Length 3 for i=0 to Sent. Length-l 4 j=l+i; 5 guessrule(i, j, XRules); 6 foreach rule in XRules 7 if rule exists in Graph table 8 add new rule to [X, i, j]; 9 Get k-best derivations of [X, i, j];
![MI&T LAB 每个结点的k-best计算 Make a heap as a priority queue For every rule push MI&T LAB 每个结点的k-best计算 Make a heap as a priority queue For every rule push](http://slidetodoc.com/presentation_image_h2/d2072b75d2f1f05c564ad5a88929c16f/image-26.jpg)
MI&T LAB 每个结点的k-best计算 Make a heap as a priority queue For every rule push its 1 st best translation into the heap While(candi. size()<k && heap. size() != 0) candi. push back(heap. top()) pop heap push neighbors of heap. top() into heap
![](http://slidetodoc.com/presentation_image_h2/d2072b75d2f1f05c564ad5a88929c16f/image-27.jpg)
![MI&T LAB IWSLT 2004测试 System Pharaoh Sub-graph pairs BLEU-4 0. 3548± 0. 0082 0. MI&T LAB IWSLT 2004测试 System Pharaoh Sub-graph pairs BLEU-4 0. 3548± 0. 0082 0.](http://slidetodoc.com/presentation_image_h2/d2072b75d2f1f05c564ad5a88929c16f/image-28.jpg)
MI&T LAB IWSLT 2004测试 System Pharaoh Sub-graph pairs BLEU-4 0. 3548± 0. 0082 0. 3714± 0. 0085
![MI&T LAB 一些翻译结果 Pharaoh They must compensate our losses Subgraph They must compensate our MI&T LAB 一些翻译结果 Pharaoh They must compensate our losses Subgraph They must compensate our](http://slidetodoc.com/presentation_image_h2/d2072b75d2f1f05c564ad5a88929c16f/image-29.jpg)
MI&T LAB 一些翻译结果 Pharaoh They must compensate our losses Subgraph They must compensate our losses Pharaoh They will be compensation for our losses Subgraph They will compensate our losses Pharaoh According to regulations, they will be our loss compensation Subgraph According to regulations, they will compensate our losses
![](http://slidetodoc.com/presentation_image_h2/d2072b75d2f1f05c564ad5a88929c16f/image-30.jpg)
![MI&T LAB 谢 谢! MI&T LAB 谢 谢!](http://slidetodoc.com/presentation_image_h2/d2072b75d2f1f05c564ad5a88929c16f/image-31.jpg)
MI&T LAB 谢 谢!
- Slides: 31