BLEU Kishore Papineni, Salim Roukos, Todd Ward and Wei-Jing Zhu. (2002) BLEU: a method for Automatic Evaluation of Machine Translation. ACL. MT ( ) MT

Similar documents
2

Rapp BLEU[10] [9] BLEU OrthoBLEU Rapp OrthoBLEU [9] OrthoBLEU OrthoBLEU ) ) ) 1) NTT Natural Language Research

f ê ê = arg max Pr(e f) (1) e M = arg max λ m h m (e, f) (2) e m=1 h m (e, f) λ m λ m BLEU [11] [12] PBMT 2 [13][14] 2.2 PBMT Hiero[9] Chiang PBMT [X

cat A



untitled


<90F496CA30305F955C8E862E6169>

FX自己アフリエイトマニュアル

FX ) 2

PDF


IPSJ-TOD

IPSJ SIG Technical Report Vol.2014-NL-219 No /12/17 1,a) Graham Neubig 1,b) Sakriani Sakti 1,c) 1,d) 1,e) 1. [23] 1(a) 1(b) [19] n-best [1] 1 N




IBM-Mode1 Q: A: cash money It is fine today 2

2 3


( ) Kevin Duh

JA2008

東海道新幹線でDS





net-h12_12.doc

B B B B B B

(資料2)第7回資料その1(ヒアリング概要)



<4D F736F F D DEC8BC A95BD90AC E A982BA81698AB A B B4790DF90AB8EBE8AB FC89408A4F816A82CC93AE8CFC82C98AD682B782E9838C837C815B D

, , ,210 9, ,

A4.pdf


Present Situation and Problems on Aseismic Design of Pile Foundation By H. Hokugo, F. Ohsugi, A. Omika, S. Nomura, Y. Fukuda Concrete Journal, Vol. 29

89-95.indd

研究報告 2 自動評価法を用いた機械翻訳の定量的評価 越前谷博 ( 北海学園大学 ) 磯崎秀樹 ( 岡山県立大学 ) 目次 1. 自動評価法とは 2. 自動評価法における動向 Workshop on Statistical Machine Translationに参加して 3. 自動評価法 :APA

,000m 7 CAT

1 1 tf-idf tf-idf i

1' ド Ui:;~


IPSJ SIG Technical Report Vol.2016-CE-137 No /12/ e β /α α β β / α A judgment method of difficulty of task for a learner using simple


untitled

シンポジウム 報告書_5.PDF


アジア言語を中心とした機械翻訳の評価 第 1 回アジア翻訳ワークショップ概要 Evaluation of Machine Translation Focusing on Asian Languages Overview of the 1st Workshop on Asian Translation

[1] [2] [3] 1 GPS 1 Twitter *1 *1 GPS [4] [5] [6] 2 [7] 1 [8] Restricted Boltzmann Machine RBM RBM


2013_autumn.indb

IPSJ SIG Technical Report An Evaluation Method for the Degree of Strain of an Action Scene Mao Kuroda, 1 Takeshi Takai 1 and Takashi Matsuyama 1

先端社会研究3号/3.金菱

  医 事 法        山口大法医 藤宮龍也


新:コミュ障レポート.pages

udc-2.dvi

0|0・・f0C0H

CVaR



2017 (413812)


1: A/B/C/D Fig. 1 Modeling Based on Difference in Agitation Method artisoc[7] A D 2017 Information Processing

ÉXÅ[ÉpÅ[ÉRÉuÉâï\éÜ

(iii) x, x N(µ, ) z = x µ () N(0, ) () 0 (y,, y 0 ) (σ = 6) *3 0 y y 2 y 3 y 4 y 5 y 6 y 7 y 8 y 9 y ( ) *4 H 0 : µ

H2-P3.indd

1

ÉgÉEÉRÉuÉâï\éÜ

1 1,200m ,482,620m 1 323,825m 1 205,988m 114,988m 60,000m 31,000m ,837m ,158,795m ,340m 1 48,006m 1 33,016m 1,125m

2

ρ(= kg m 3 ), g h P 0 C () [1] 1.3 SI Pa hpa h 100 ( : 100 ) 1m 2 1N 1Pa 1N 1kg 1m s 2 Pa hpa mb hpa 1mm 1mmHg hpa 1mmHg =


Abstract :

3 ( 9 ) ( 13 ) ( ) 4 ( ) (3379 ) ( ) 2 ( ) 5 33 ( 3 ) ( ) 6 10 () 7 ( 4 ) ( ) ( ) 8 3() 2 ( ) 9 81

Fig. 1 Schematic construction of a PWS vehicle Fig. 2 Main power circuit of an inverter system for two motors drive

[2] , [3] 2. 2 [4] 2. 3 BABOK BABOK(Business Analysis Body of Knowledge) BABOK IIBA(International Institute of Business Analysis) BABOK 7

23 The Study of support narrowing down goods on electronic commerce sites

公共建築物構造設計の用途係数基準

Microsoft Word - 倫理 第40,43,45,46講 テキスト.docx


03山本雅子.indd

Pari-gp /7/5 1 Pari-gp 3 pq



dy + P (x)y = Q(x) (1) dx dy dx = P (x)y + Q(x) P (x), Q(x) dy y dx Q(x) 0 homogeneous dy dx = P (x)y 1 y dy = P (x) dx log y = P (x) dx + C y = C exp

131WA (ai).pdf


kut-paper-template.dvi

lagged behind social progress. During the wartime Chonaikai did cooperate with military activities. But it was not Chonaikai alone that cooperated. Al

IPSJ SIG Technical Report Pitman-Yor 1 1 Pitman-Yor n-gram A proposal of the melody generation method using hierarchical pitman-yor language model Aki

論文9.indd

Core Ethics Vol.


- 2 -

untitled

(3)

gc_ indd

Transcription:

4. BLEU @NICT mutiyama@nict.go.jp 1

BLEU Kishore Papineni, Salim Roukos, Todd Ward and Wei-Jing Zhu. (2002) BLEU: a method for Automatic Evaluation of Machine Translation. ACL. MT ( ) MT ( ) BLEU 2

BLEU BLEU MT NTCIR-7 3

ngram 1 MT ngram MT 1. 1 It is a guide to action 2 which 3 ensures that the military 4always obyes the 5 commands 6 of the party. 2. It is to ensure the troops forever hearing the activity guidebook that party direct 1. 1 It is a guide to action that 3 ensures that the military will forever heed Party 5 commands. 2. It is the guiding principle 2 which guarantees the military forces 4 always being under the command 6 of the party. 3. It is the practical guide for the army 4 always to heed the directions 6 of the party. 4

1. ngram MT 2. 1. ngram 2. 5

ngram ngram = ngram MT ngram MT: the the the the the the the Ref1: The cat is on the mat. Ref2: There is a cat on the mat. MT the the Ref1 Ref2 1gram = 7 7 6

ngram ngram ngram P n = MT ngram MT: the the the the the the the Ref1: The cat is on the mat. Ref2: There is a cat on the mat. P 1 = 2 7 P 2 = 0 7

MT1: P 1 = 17 18 = 0.94, P 2 = 10 17 = 0.59 MT2: P 1 = 8 14 = 0.57, P 2 = 1 13 = 0.08 MT 1. 1 It is a guide to action 2 which 3 ensures that the military 4always obyes the 5 commands 6 of the party. 2. It is to ensure the troops forever hearing the activity guidebook that party direct 1. 1 It is a guide to action that 3 ensures that the military will forever heed Party 5 commands. 2. It is the guiding principle 2 which guarantees the military forces 4 always being under the command 6 of the party. 3. It is the practical guide for the army 4 always to heed the directions 6 of the party. 8

ngram P n = MT MT ngram ngram MT MT ngram ngram MT ngram MT ngram MT 9

ngram N 1 n=1 N log P n P n = ngram N = ngram ( 4 ) ngram 10

ngram P n MT ngram ngram MT P n MT ngram P 1 = 2/2, P 2 = 1/1 MT : of the 1: It is a guide to action that ensures that the military will forever heed Party commands. 2: It is the guiding principle which guarantees the military forces always being under the command of the party. 3: It is the practical guide for the army always to heed the directions of the party. 11

MT MT ( ) c = MT MT r = MT BP (brevity penalty) BP = 1 if c r exp(1 r/c) if c < r MT BP = 1 ( ) MT < BP < 1 12

BLEU BLEU = BP exp( N 1 n=1 N log P n) MT ngram BLEU BLEU... MT 13

BLEU 10 ( ) 2 10 ( ) 500 50 5 (S1,S2,S3,H1,H2) H1 H2 H1 H2 1( ) 5( ) 2 0 14

(paired t-test) u i s r(u, i, s) s r(u, i, s ) s s d(u, i, s, s ) = r(u, i, s) r(u, i, s ) m(s, s ) = u,i d(u, i, s, s ) v(s, s ) = 1 n u,i (d(u, i, s, s ) m(s, s )) 2 n = u,i 1. s s m(s, s ) = 0 0 t = m(s, s ) v(s,s ) n 1 t 15

BLEU BLEU 2 S1 0.0527 0 0 S2 0.0829 0.326 0.551 S3 0.0930 0.44 0.691 H1 0.1934 2.265 2.574 H2 0.2571 2.8 2.612 2 S1 0 S2 m(s1, S2) S3 = S2+m(S3, S2) 16

BLEU 1 0.8 Normalized score 0.6 0.4 0.2 0 Bilingual Monolingual BLEU S1 S2 S3 H1 H2 0-1 BLEU 2 BLEU S3 H1 H1 H2 S2 S3 17

BLEU 3 2.5 2 Monolingual Judgment 1.5 1 0.5 0 monolingual.txt f(x) -0.5 0 0.05 0.1 0.15 0.2 0.25 0.3 BLEU BLEU 18

BLEU 3 2.5 2 Bilingual Judgment 1.5 1 0.5 0 bilingual.txt f(x) -0.5 0 0.05 0.1 0.15 0.2 0.25 0.3 BLEU BLEU 2 19

BLEU MT BLEU BLEU ngram 20