Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
An-Operation-Network-for-Abstractive-Sentence-Compression.pdf
Search
MARUYAMA
July 25, 2018
0
86
An-Operation-Network-for-Abstractive-Sentence-Compression.pdf
MARUYAMA
July 25, 2018
Tweet
Share
More Decks by MARUYAMA
See All by MARUYAMA
vampire.pdf
tmaru0204
0
130
Misspelling_Oblivious_Word_Embedding.pdf
tmaru0204
0
130
Simple_Unsupervised_Summarization_by_Contextual_Matching.pdf
tmaru0204
0
130
Controlling_Text_Complexity_in_Neural_Machine_Translation.pdf
tmaru0204
0
120
20191028_literature-review.pdf
tmaru0204
0
120
Hint-Based_Training_for_Non-Autoregressive_Machine_Translation.pdf
tmaru0204
0
100
Soft_Contextual_Data_Augmentation_for_Neural_Machine_Translation_.pdf
tmaru0204
0
120
An_Embarrassingly_Simple_Approach_for_Transfer_Learning_from_Pretrained_Language_Models_.pdf
tmaru0204
0
110
Addressing_Trobulesome_Words_in_Neural_Machine_Translation.pdf
tmaru0204
0
110
Featured
See All Featured
The Invisible Customer
myddelton
114
12k
Typedesign – Prime Four
hannesfritz
36
2.1k
GitHub's CSS Performance
jonrohan
1025
450k
Reflections from 52 weeks, 52 projects
jeffersonlam
345
19k
How to train your dragon (web standard)
notwaldorf
73
5.2k
Building an army of robots
kneath
300
41k
Fashionably flexible responsive web design (full day workshop)
malarkey
398
65k
Visualizing Your Data: Incorporating Mongo into Loggly Infrastructure
mongodb
34
8.9k
KATA
mclloyd
15
12k
Web Components: a chance to create the future
zenorocha
305
41k
Designing Experiences People Love
moore
136
23k
How STYLIGHT went responsive
nonsquared
92
4.8k
Transcript
An Operation Network for Abstractive Sentence Compression Naitong Yu, Jie
Zhang, Minlie Huang, Xiaoyan Zhu The 27th International Conference on Computational Linguistics (COLING 2018) Nagaoka University of Technology Takumi Maruyama Literature review:
Introduction Ø %*, • % (" %+ Ø 2
• Delete-based approach • Generate-based approach Ø 0/- • %*, )&.$ • Delete-based approach Generate-based approach # State-of-the-art'!
Introduction Ø Delete-based approach • )+59G%?7”” • 5BC””FE6 D2
• “” & $(5BC@ Ø Generate-based approach • “=;””>A”, “.<”, “H)”& • ,4*”=;”/0 8: Delete-based approachGenerate-based approach /' “=;”-) "!#31
Baselines Ø Seq2seq (generate-only model)
Baselines Ø Pointer-Generator (copy-and-generate model)
Method Ø Operation Network
Method Ø Delete decoder • • !" ∈ $, &
'( : *+,ℎ.//01 '2320, 4( : 4512062 704258 0(6( ): 6( 0;<0//.1=
Method Ø Copy-Generate decoder • Generate probability Generate modeCopy mode
- Generate mode - Copy mode attention distribution • Final probability distribution
Method Ø Copy-Generate decoder •
Dataset Ø Toutanova et al. (2016) • Business letters, news
journals, technical documents • Training set: 21, 145 pairs Validation set: 1,908 pairs Test set: 3,370 pairs
Evaluation Metrics Ø Automatic evaluation • Compression Ratio • ROUGE
(ROUGE-1, ROUGE-2, ROUGE-L) • BLEU Ø Manual evaluation • Grammaticality • Non-Redundancy
Results
Results
Conclusion Ø Delete-based approachGenerate-based approach Ø Delete
Ø Abstractive sentence compressionSOTA