文献紹介: Pre-Translation for Neural Machine Translation

F637b583c221c132af26c91cb3dba0ca?s=47 Yumeto Inaoka
June 21, 2017
Technology
0
59

文献紹介: Pre-Translation for Neural Machine Translation

2017/06/21の文献紹介で発表

F637b583c221c132af26c91cb3dba0ca?s=128

Yumeto Inaoka

June 21, 2017
Tweet

More Decks by Yumeto Inaoka

See All by Yumeto Inaoka
F637b583c221c132af26c91cb3dba0ca?s=48 yumeto
1
18
F637b583c221c132af26c91cb3dba0ca?s=48 yumeto
0
25
F637b583c221c132af26c91cb3dba0ca?s=48 yumeto
0
38
F637b583c221c132af26c91cb3dba0ca?s=48 yumeto
0
34
F637b583c221c132af26c91cb3dba0ca?s=48 yumeto
0
23
F637b583c221c132af26c91cb3dba0ca?s=48 yumeto
0
56
F637b583c221c132af26c91cb3dba0ca?s=48 yumeto
0
86
F637b583c221c132af26c91cb3dba0ca?s=48 yumeto
0
61
F637b583c221c132af26c91cb3dba0ca?s=48 yumeto
0
63

Other Decks in Technology

See All in Technology
69b4dadef85efe143ac825b62d36ed7c?s=48 fruitriin
2
260
0251532f4fb413ea1a026efe6eb16b87?s=48 stajima
0
210
75fd80a8ace6ab292a7de563e018ae11?s=48 takashi_toyosaki
0
130
A500dfe11c36aa2912eaf33176fe129f?s=48 oyakata2438
3
410
961126d9892ae4df45249529a0721571?s=48 sugamasao
0
310
9ea422a61c1a69c888786830eee3dbe6?s=48 osonoi
0
130
13d936e697fe0f4fa96f926d0a712f6c?s=48 sansanbuildersbox
0
130
69f5ab96892df4d8cbe32189f2ed2d9d?s=48 heyinc
1
2.6k
A38e9ebe0e62fd6753f555c005ea6551?s=48 tarukosu
0
490
8db231d3fe08b46242f6e0e45c95eee1?s=48 darkomesaros
0
110
53850955f15249a1a9dc49df6113e400?s=48 line_developers
0
140
88f4e84b94fe07cddbd9e6479d689192?s=48 soudai
0
160

Featured

See All Featured
78b475797a14c84799063c7cd073962f?s=48 holman
446
130k
245cee81a9c424266e5e401d844ea881?s=48 lara
587
60k
E14f55d3f939977cecbf51b64ff6f861?s=48 keithpitt
398
20k
168ccec72eee0530b818d44f3fedaacf?s=48 shlominoach
174
7.1k
9128d500301ae51524e887bb680f471d?s=48 caitiem20
300
16k
Be9caeb9d4ef9944d151af909063ed6e?s=48 samlambert
236
9.7k
D67fff74178013024d833b3eec6cf27f?s=48 lynnandtonic
264
16k
F68cb25ae3e5c2106997454b13b7737d?s=48 brad_frost
154
6.1k
5f2da528927a2ec9ba4fec2069cbc958?s=48 kneath
293
38k
4262b9cfacfb6b2c77f23e1d0310cd3e?s=48 myddelton
104
10k
5f2da528927a2ec9ba4fec2069cbc958?s=48 kneath
218
14k
C1daf20e5c49ff910745198ef9869ac2?s=48 hatefulcrawdad
257
16k

Transcript

  1. 1SF5SBOTMBUJPOGPS
 /FVSBM.BDIJOF5SBOTMBUJPO จݙ঺հ   ࣗવݴޠॲཧݚڀࣨɹҴԬເਓ +BO/JFIVFT &VOBI$IP 5IBOI-F)BBOE"MFY8BJCFM 1SPDFFEJOHTPG$0-*/(

    QQ

  2. ֓ཁ ˗ /.5͸࣌ʹશ͘ҟͳΔҙຯͷ຋༁Λੜ੒
 ˡ·Εͳ୯ޠ͕͋Δ৔߹ʹݦஶ ˗ 1#.5ʹΑΔࣄલ຋༁Ͱ໨తݴޠʹ຋༁
 ˡ্ه໰୊Λճආ ˗ /.5ʹೖྗจͱࣄલ຋༁ͷ݁ՌΛೖྗ
 ˠ#-&6είΞͷ޲্

    

  3. /.5ͱ1#.5 ˗ /.5
 ·Εͳ୯ޠͷ຋༁ʹࣦഊ
 ྲྀெͰจ๏తʹਖ਼͍͠຋༁Λੜ੒ ˗ 1#.5
 ·Εͳ୯ޠͰ΋຋༁͕Մೳ 

  4. ύΠϓϥΠϯʹΑΔ݁߹ ˗ 1#.5Ͱೖྗݴޠ͔Β໨తݴޠF`ʹ຋༁ ˗ ্ͷग़ྗΛ/.5ͰΑΓྑ͍໨తจFʹ຋༁ ˗ 1#.5͕/.5ͰϦΧόʔͰ͖ͳ͍ΤϥʔΛ
 Ҿ͖ى͜͢Մೳੑ 

  5. /.5΁ͷࠞ߹ೖྗ ˗ /.5ʹ1#.5ͷग़ྗͱݪจΛೖྗ ˗ 1#.5ͷग़ྗޠኮͱݪจͷޠኮͷॏෳΛ๷ࢭ ˗ ͦΕͧΕͷೖྗΛ࿈݁ DPODBUFOBUJPO 

  6. ࣮ݧ ˗ 8.5ͷӳಠχϡʔε຋༁λεΫͰ෼ੳ ˗ #-&6Ͱ຋༁඼࣭Λ෼ੳ ˗ ୯ޠͷස౓ʹԠͯ͡ύϑΥʔϚϯεΛ෼ੳ ˗ 1#.5ͷ຋༁඼࣭ʹΑΔӨڹΛ෼ੳ͢ΔͨΊ
 ϕʔεϥΠϯγεςϜͱߴ౓ͳϞσϧΛ࢖༻

    ˗ /.5͸ϞσϧͷγεςϜͱϞσϧͷ
 ΞϯαϯϒϧγεςϜΛ࢖༻ 

  7. ࣮ݧ ˗ 1#.5
 5FSFTB)FSSNBOO +BO/JFIVFT BOE"MFY8BJCFM
 $PNCJOJOH8PSE3FPSEFSJOH.FUIPETPOEJ⒎FSFOU-JOHVJTUJD"CTUSBDUJPO -FWFMTGPS4UBUJTUJDBM.BDIJOF5SBOTMBUJPO
 *O1SPDFFEJOHTPGUIF4FWFOUI8PSLTIPQPO4ZOUBY 4FNBOUJDTBOE

    4USVDUVSFJO4UBUJTUJDBM5SBOTMBUJPO "MUBOUB (FPSHJB 64" ˗ /.5
 3JDP4FOOSJDI #BSSZ)BEEPX BOE"MFYBOESB#JSDI
 /FVSBMNBDIJOFUSBOTMBUJPOPGSBSFXPSETXJUITVCXPSEVOJUT
 *O1SPDFFEJOHTPGUIFUI"OOVBM.FFUJOHPGUIF"TTPDJBUJPOGPS $PNQVUBUJPOBM-JOHVJTUJDT "$- "VHVTU  #FSMJO (FSNBOZ  7PMVNF-POH1BQFST 

  8. ݁Ռ #-&6είΞ ˗ ύΠϓϥΠϯ͸/.5ΑΓੑೳ͕ྼΔ
 ˠ1#.5ͷΤϥʔΛճ෮Ͱ͖ͳ͍ͨΊ ˗ ࠞ߹Ͱ͸ੑೳ޲্ 

  9. ݁Ռ ྫจ ˗ 1#.5Ͱ͸୯ޠͷ༁͸ਖ਼͍͕͠จ๏ʹΤϥʔ ˗ /.5͸୯ޠͷ༁ͰΤϥʔ ˗ ఏҊख๏Ͱ͸྆ํ͕վળ 

  10. ݁Ռ ྫจ ˗ ೖྗจʹ௿ස౓ޠSJPUؚ͕·ΕΔ ˗ 1#.5Ͱ͸ਖ਼͘͠"VGSVISʹ༁͞ΕΔ ˗ /.5͸ͦͷ··ग़ྗ ˗ ఏҊख๏Ͱਖ਼͘͠຋༁

    

  11. ·ͱΊ ˗ 1#.5ͱ/.5Λ૊Έ߹Θͤͨख๏ΛఏҊ ˗ ྆ϞσϧΛ௚ྻͰ૊Έ߹Θͤͯ΋༏Εͨ຋༁͸ ੜ੒Ͱ͖ͳ͍͕ೖྗΛࠞ߹͢ΔͱվળͰ͖Δ ˗ 1#.5ͷग़ྗͱݪจΛγϯϓϧʹ࿈݁ DPODBUFOBUJPO ͢Δ͜ͱͰ඼࣭͕޲্

    ˗ ݸʑΛվળ͢Δ͜ͱͰγεςϜશମͷ
 ύϑΥʔϚϯε͕޲্