Entry Liu:2014:DTL from talip.bib
Last update: Sun Oct 15 02:55:04 MDT 2017
Top |
Symbols |
Numbers |
Math |
A |
B |
C |
D |
E |
F |
G |
H |
I |
J |
K |
L |
M |
N |
O |
P |
Q |
R |
S |
T |
U |
V |
W |
X |
Y |
Z
BibTeX entry
@Article{Liu:2014:DTL,
author = "Lemao Liu and Tiejun Zhao and Taro Watanabe and
Hailong Cao and Conghui Zhu",
title = "Discriminative Training for Log-Linear Based {SMT}:
Global or Local Methods",
journal = j-TALIP,
volume = "13",
number = "4",
pages = "17:1--17:??",
month = dec,
year = "2014",
CODEN = "????",
DOI = "https://doi.org/10.1145/2637478",
ISSN = "1530-0226 (print), 1558-3430 (electronic)",
ISSN-L = "1530-0226",
bibdate = "Wed Jan 7 15:23:49 MST 2015",
bibsource = "http://portal.acm.org/;
http://www.math.utah.edu/pub/tex/bib/talip.bib",
abstract = "In statistical machine translation, the standard
methods such as MERT tune a single weight with regard
to a given development data. However, these methods
suffer from two problems due to the diversity and
uneven distribution of source sentences. First, their
performance is highly dependent on the choice of a
development set, which may lead to an unstable
performance for testing. Second, the sentence level
translation quality is not assured since tuning is
performed on the document level rather than on sentence
level. In contrast with the standard global training in
which a single weight is learned, we propose novel
local training methods to address these two problems.
We perform training and testing in one step by locally
learning the sentence-wise weight for each input
sentence. Since the time of each tuning step is
unnegligible and learning sentence-wise weights for the
entire test set means many passes of tuning, it is a
great challenge for the efficiency of local training.
We propose an efficient two-phase method to put the
local training into practice by employing the
ultraconservative update. On NIST Chinese-to-English
translation tasks with both medium and large scales of
training data, our local training methods significantly
outperform standard methods with the maximal
improvements up to 2.0 BLEU points, meanwhile their
efficiency is comparable to that of the standard
methods.",
acknowledgement = ack-nhfb,
articleno = "17",
fjournal = "ACM Transactions on Asian Language Information
Processing",
journal-URL = "http://portal.acm.org/browse_dl.cfm?idx=J820",
}
Related entries
- address,
6(3)11,
7(1)2,
7(1)3,
8(2)6,
8(4)15,
8(4)18,
8(4)19,
9(1)3,
10(3)13,
10(4)21,
11(2)6,
11(4)13,
11(4)17,
13(1)1,
13(2)8
- based,
1(3)269,
2(4)301,
3(2)113,
4(3)357,
4(4)377,
5(2)121,
5(2)165,
5(3)185,
6(2)8,
6(3)9,
6(3)11,
6(4)3,
7(2)6,
7(3)10,
7(4)12,
8(1)4,
8(3)11,
8(3)12,
8(4)17,
8(4)19,
9(1)2,
9(2)5,
9(2)6,
9(2)7,
9(3)11,
10(1)4,
10(1)6,
10(2)7,
10(3)14,
10(3)16,
11(2)6,
11(3)8,
11(3)11,
11(4)14,
11(4)15,
11(4)16,
11(4)17,
11(4)18,
12(1)1,
12(1)2,
12(1)4,
12(3)9,
12(3)10,
12(3)11,
13(1)2,
13(1)3,
13(1)4,
13(2)10,
13(3)13,
13(4)18
- BLEU,
10(4)18,
12(4)17
- both,
6(2)6,
6(2)7,
6(3)10,
7(1)1,
7(1)2,
7(1)3,
7(3)8,
8(2)7,
8(3)11,
8(3)12,
9(1)2,
9(1)4,
9(2)5,
10(1)2,
10(1)4,
10(1)6,
10(2)10,
10(3)13,
10(3)15,
10(4)20,
10(4)21,
11(2)4,
11(2)6,
11(2)7,
11(3)9,
11(4)18,
12(2)5,
12(2)7,
12(4)17,
13(2)9
- challenge,
4(3)243,
6(3)10,
6(4)3,
8(1)4,
8(4)14,
8(4)18,
9(1)3,
10(3)12,
13(1)1,
13(1)4,
13(2)8,
13(3)12
- Chinese-to-English,
7(1)1,
8(1)4,
8(2)6
- choice,
9(2)6,
11(4)13,
12(4)14,
12(4)15
- comparable,
6(3)11,
10(4)19,
11(2)6,
12(3)9,
12(3)12
- contrast,
10(3)14,
13(1)2
- data,
2(2)143,
6(1)z,
6(1)z-1,
6(2)7,
6(3)11,
7(1)3,
7(3)9,
7(4)13,
8(1)3,
8(2)7,
8(3)10,
8(3)11,
8(3)12,
8(4)16,
8(4)18,
9(2)6,
10(2)7,
10(3)12,
10(4)20,
11(2)4,
11(3)10,
11(3)11,
11(4)13,
11(4)14,
11(4)18,
12(1)1,
12(2)7,
12(3)9,
13(1)2,
13(1)3,
13(1)4,
13(4)16,
13(4)18
- dependent,
9(2)7
- development,
2(2)101,
2(3)290,
5(2)146,
6(2)6,
9(1)4,
10(2)10,
12(2)6
- discriminative,
1(1)34,
3(2)128,
5(4)413,
8(2)8,
8(4)15,
10(4)17,
11(2)7
- distribution,
10(1)6,
11(2)4,
11(2)5,
11(3)9,
11(4)17
- document,
1(3)225,
2(1)1,
3(2)128,
4(3)340,
5(2)121,
6(4)2,
7(1)3,
7(2)7,
7(3)8,
8(1)1,
8(1)2,
8(1)3,
8(3)11,
8(3)12,
8(4)19,
9(3)12,
11(2)5,
12(1)3,
13(1)1,
13(2)6,
13(2)7,
13(2)9,
13(3)13
- due,
5(2)121,
6(2)8,
7(2)7,
9(1)1,
9(1)2,
9(4)15,
10(2)10,
10(3)12,
10(4)21,
11(2)7,
12(4)14,
13(1)4
- each,
5(2)121,
5(2)146,
5(2)165,
7(3)8,
7(3)10,
7(4)11,
7(4)13,
8(4)16,
8(4)18,
9(1)2,
10(1)2,
10(2)7,
10(2)9,
10(4)20,
12(1)2,
12(1)3,
12(3)9,
12(3)10,
13(1)2,
13(2)9,
13(3)13,
13(3)14
- efficiency,
5(2)89,
6(2)8,
6(3)9,
7(3)8,
13(4)18
- efficient,
2(1)63,
4(4)377,
11(2)4,
12(2)6,
12(3)12,
13(1)4,
13(2)8,
13(4)18
- employing,
10(3)15,
11(2)4
- English, Chinese-to-,
7(1)1,
8(1)4,
8(2)6
- entire,
8(2)8,
9(1)2,
11(3)10
- first,
5(2)165,
6(2)6,
6(4)3,
7(1)1,
7(3)8,
7(3)10,
8(2)7,
8(3)10,
8(3)11,
8(4)19,
9(3)10,
9(3)11,
10(3)13,
11(2)6,
11(2)7,
11(3)8,
11(3)9,
11(3)11,
12(1)3,
12(1)4,
12(2)5,
12(3)10,
12(4)17,
13(2)7,
13(2)9,
13(3)13
- given,
5(2)89,
5(2)165,
7(3)10,
7(4)13,
8(1)2,
9(1)1,
9(2)7,
11(2)5,
11(3)8,
11(3)11,
11(4)13,
12(4)14,
13(1)2
- global,
7(4)12,
8(2)8,
8(4)15,
12(2)5
- great,
10(4)21,
11(1)1,
12(4)15
- highly,
5(2)121,
6(3)9,
6(4)2,
7(3)9,
8(4)18,
10(2)8,
10(4)20,
11(2)4,
12(2)5,
13(1)4
- however,
6(2)8,
7(2)7,
7(3)8,
7(3)10,
7(4)11,
7(4)12,
7(4)13,
8(3)10,
9(1)1,
9(3)12,
10(3)12,
10(4)18,
11(2)6,
11(2)7,
12(1)2,
12(2)7,
12(3)10,
12(3)11,
12(4)14,
13(3)11
- improvement,
4(3)280,
5(2)121,
6(2)7,
7(1)2,
7(1)3,
7(3)8,
7(3)10,
8(1)4,
8(3)10,
8(4)15,
8(4)16,
9(1)3,
9(3)11,
10(2)8,
11(4)17,
11(4)18,
12(1)1,
12(2)7,
12(3)11,
13(3)12,
13(3)14,
13(4)16
- input,
5(2)165,
8(2)7,
9(1)1,
9(1)3,
9(2)6,
10(1)6,
10(4)20,
12(1)2,
12(1)4,
12(2)6,
13(1)4,
13(2)8,
13(3)12
- large,
6(2)8,
6(3)9,
6(3)11,
7(3)8,
7(3)9,
7(4)11,
8(1)2,
8(3)12,
8(4)18,
9(4)14,
10(2)7,
10(4)20,
10(4)21,
11(1)2,
11(3)8,
12(2)6,
12(3)9,
13(1)3,
13(2)7,
13(2)8,
13(3)13,
13(4)18
- lead,
8(4)16,
9(1)4,
12(1)1
- learned,
4(2)78,
13(4)16
- learning,
3(2)159,
5(1)61,
5(2)121,
5(4)413,
6(2)6,
6(4)1,
6(4)3,
7(2)7,
7(3)9,
8(3)10,
8(4)15,
9(1)3,
9(2)5,
10(1)5,
10(2)10,
10(3)16,
10(4)20,
10(4)21,
11(1)3,
11(4)14,
11(4)16,
12(1)1,
12(1)3,
12(2)5,
12(2)7,
12(4)15,
13(1)2,
13(1)3,
13(2)9,
13(4)16
- level,
5(2)89,
8(4)15,
9(3)11,
10(3)12,
11(3)8,
12(3)9,
13(1)3,
13(1)4,
13(3)12
- local,
7(4)12,
8(1)4,
8(4)15,
9(1)1
- machine,
4(1)18,
4(4)377,
5(2)89,
5(3)185,
6(2)6,
6(4)2,
6(4)3,
7(1)1,
7(2)5,
7(2)7,
7(3)9,
7(3)10,
8(1)4,
8(2)5,
8(2)6,
8(2)7,
8(2)8,
8(2)9,
8(3)10,
8(4)15,
9(1)3,
9(4)13,
10(1)2,
10(1)5,
10(3)16,
10(4)18,
10(4)20,
11(1)2,
11(3)8,
11(4)14,
11(4)16,
12(3)9,
12(3)12,
12(4)14,
12(4)16,
12(4)17,
13(1)2,
13(3)11,
13(4)16
- many,
5(2)89,
5(2)121,
7(2)7,
7(4)11,
7(4)12,
8(4)17,
9(1)1,
9(4)13,
10(1)2,
10(1)5,
10(4)17,
10(4)19,
11(1)2,
12(1)2,
12(2)6,
12(3)9,
12(4)16
- may,
8(3)11,
8(4)16,
9(1)1,
9(4)13,
12(1)4,
12(3)10
- mean,
6(3)11,
7(2)6,
9(3)11,
10(1)3,
12(3)11,
13(2)6
- NIST,
8(2)6
- not,
5(2)89,
5(2)121,
6(3)10,
7(2)7,
7(3)8,
7(3)10,
7(4)11,
7(4)12,
7(4)13,
8(1)3,
8(2)7,
8(2)8,
8(2)9,
8(4)17,
9(2)6,
9(2)7,
9(4)14,
10(1)4,
10(1)5,
10(2)8,
10(2)10,
10(4)18,
11(1)2,
11(3)10,
12(1)2,
12(1)3,
12(2)7,
12(3)9,
13(2)7,
13(2)9,
13(4)18
- novel,
4(3)243,
5(2)89,
5(2)165,
6(2)8,
6(4)2,
8(1)3,
8(3)12,
10(2)7,
10(3)14,
10(4)19,
10(4)20,
11(1)3,
11(2)4,
11(3)11,
11(4)15,
11(4)17,
12(3)11,
12(3)12,
13(2)8,
13(3)13
- one,
5(2)89,
5(2)121,
6(2)6,
6(3)9,
6(4)3,
7(3)8,
7(3)9,
7(4)11,
7(4)13,
8(2)9,
8(4)16,
8(4)17,
9(1)1,
9(2)5,
9(2)7,
9(3)12,
9(4)14,
10(1)5,
10(3)12,
10(3)13,
10(4)19,
11(2)4,
11(2)6,
11(2)7,
11(4)14,
12(1)1,
12(1)2,
12(2)5,
12(2)7,
12(3)11,
12(4)16,
13(1)4,
13(2)10,
13(4)18
- outperform,
6(4)3,
7(2)6,
7(3)10,
7(4)13,
8(2)6,
9(2)5,
9(2)6,
10(3)15,
11(2)7,
11(3)8,
11(4)14,
12(2)5,
13(3)13
- pass,
8(1)4
- perform,
7(1)1,
9(3)12,
9(4)14,
11(2)4,
11(3)11,
11(4)16,
12(1)2
- performance,
5(2)121,
5(2)165,
6(2)8,
6(3)9,
6(4)1,
6(4)3,
7(1)1,
7(1)2,
7(2)5,
7(2)6,
7(2)7,
7(3)9,
7(3)10,
7(4)13,
8(1)2,
8(1)3,
8(2)7,
8(2)8,
8(2)9,
8(3)10,
8(4)16,
8(4)17,
8(4)18,
9(1)2,
9(1)4,
9(2)5,
9(2)6,
9(3)11,
9(3)12,
9(4)14,
10(2)8,
10(3)13,
10(3)14,
11(2)7,
11(3)10,
11(3)11,
11(4)14,
11(4)15,
11(4)17,
12(1)2,
12(3)9,
12(3)11,
12(4)14,
12(4)15,
12(4)16,
13(1)3,
13(1)4,
13(2)7,
13(2)9,
13(4)16
- performed,
7(3)10,
8(1)2,
9(3)11,
10(1)2,
12(2)7,
13(2)6
- phase, two-,
12(1)3
- point,
6(2)8,
8(4)14,
10(1)6,
12(3)9,
12(4)16,
13(2)6
- practice,
5(2)146,
12(1)2
- problem,
6(2)7,
6(3)9,
6(3)11,
6(4)1,
7(1)2,
7(2)7,
7(3)10,
8(1)2,
8(2)9,
8(3)10,
8(4)19,
9(1)1,
9(1)3,
9(2)5,
9(4)13,
10(1)2,
10(1)4,
10(3)14,
10(3)16,
10(4)21,
11(3)8,
11(3)11,
11(4)17,
11(4)18,
12(1)2,
12(1)3,
12(2)7,
12(3)10,
12(3)12,
12(4)16,
13(2)8,
13(3)14
- propose,
5(2)89,
6(2)6,
6(2)8,
6(3)11,
7(3)8,
7(3)10,
7(4)12,
8(1)2,
8(1)4,
8(2)9,
8(4)19,
9(2)7,
9(4)13,
10(2)10,
10(3)12,
10(3)15,
10(4)17,
10(4)20,
11(2)7,
11(3)9,
11(3)11,
11(4)15,
11(4)16,
11(4)18,
12(1)1,
12(1)2,
12(1)3,
12(1)4,
12(2)5,
12(2)6,
12(3)9,
12(3)10,
12(3)12,
12(4)16,
13(1)2,
13(1)3,
13(2)8,
13(2)9,
13(3)12,
13(3)13,
13(4)18
- put,
8(4)17
- quality,
8(3)12,
9(4)13,
10(4)17,
11(3)8,
12(4)14,
12(4)15,
13(1)1
- rather,
11(3)9
- regard,
7(2)6,
9(2)5,
11(1)1,
11(4)16
- second,
5(2)89,
8(1)4,
8(2)7,
8(4)19,
11(1)3,
11(2)6,
11(3)8,
11(3)9,
11(3)11,
12(1)3,
13(2)7
- sentence,
1(3)173,
3(2)146,
4(3)321,
4(4)377,
5(2)121,
5(2)146,
5(2)165,
7(1)3,
7(2)6,
7(4)13,
8(1)3,
8(2)8,
9(1)2,
9(2)6,
10(4)21,
11(1)3,
11(2)5,
11(2)6,
11(3)8,
11(3)10,
11(3)11,
12(1)2,
12(1)3,
12(2)7,
12(4)14,
12(4)17,
13(1)2,
13(3)11
- set,
1(3)269,
5(2)121,
6(1)z,
7(1)3,
7(3)8,
7(4)11,
7(4)13,
8(3)10,
8(3)12,
8(4)15,
9(1)1,
9(1)3,
9(2)5,
10(1)4,
10(2)8,
10(4)20,
11(2)5,
11(2)7,
11(3)10,
11(3)11,
11(4)13,
11(4)14,
12(1)2,
12(1)4,
12(3)9,
13(2)8,
13(2)9,
13(3)12,
13(3)13
- significantly,
7(1)1,
7(4)13,
8(2)6,
8(4)15,
8(4)18,
9(1)2,
9(2)6,
9(3)12,
11(2)6,
11(4)13,
11(4)17,
12(4)16,
13(2)6
- since,
5(2)89,
5(2)165,
8(2)9,
8(4)16,
8(4)18,
10(1)4,
10(1)5,
10(4)19,
10(4)21,
11(2)7,
11(4)18,
12(1)2,
12(4)16
- single,
7(3)8,
8(3)11,
9(3)12,
11(2)5,
13(3)14
- SMT,
7(1)1,
8(1)4,
8(2)6,
10(4)17,
10(4)18,
11(3)8,
12(3)12,
12(4)17
- source,
5(2)121,
6(3)11,
8(2)6,
8(2)7,
8(2)8,
8(4)17,
9(1)1,
9(4)15,
10(3)12,
10(4)17,
10(4)19,
11(4)18,
12(4)15,
13(1)1,
13(1)2,
13(3)12
- standard,
6(3)10,
7(4)11,
8(4)14,
8(4)15,
8(4)18,
9(1)1,
9(2)5,
9(2)6,
10(3)12,
11(2)5,
11(4)18,
13(2)7,
13(3)13
- statistical,
1(1)3,
3(2)87,
3(4)243,
5(2)121,
5(4)323,
5(4)360,
6(1)z-4,
7(1)1,
8(1)2,
8(1)4,
8(2)6,
8(2)7,
8(2)8,
8(2)9,
8(3)10,
8(4)15,
8(4)19,
9(2)6,
9(2)7,
9(3)11,
10(4)18,
11(2)6,
11(2)7,
11(3)8,
11(4)15,
12(1)1,
12(3)12,
12(4)14,
12(4)16,
12(4)17,
13(1)2,
13(1)3,
13(1)4
- step,
7(3)8,
8(2)8,
8(4)16,
9(3)12,
10(1)5,
10(3)12,
11(2)6,
12(1)2,
12(2)7,
13(3)14
- such,
7(2)7,
7(3)8,
7(3)10,
7(4)12,
8(2)8,
8(3)10,
8(3)11,
8(3)12,
8(4)14,
8(4)16,
8(4)17,
9(1)1,
9(3)12,
9(4)13,
9(4)15,
10(1)5,
10(2)8,
10(3)12,
10(4)21,
11(1)2,
11(2)5,
11(2)7,
11(3)8,
11(3)10,
11(3)11,
11(4)13,
11(4)16,
11(4)17,
11(4)18,
12(1)1,
12(1)2,
12(2)6,
12(3)10,
12(3)11,
12(4)14,
12(4)17,
13(1)1,
13(3)12
- task,
2(1)49,
5(2)89,
5(2)121,
6(2)7,
6(3)11,
6(4)1,
6(4)3,
7(1)1,
7(1)2,
7(2)7,
7(3)10,
7(4)13,
8(1)4,
8(2)7,
8(4)15,
8(4)16,
9(1)4,
9(2)6,
9(3)10,
9(4)14,
9(4)15,
10(1)5,
10(3)14,
10(4)18,
10(4)20,
10(4)21,
11(1)2,
11(3)8,
11(3)11,
11(4)13,
11(4)14,
11(4)17,
11(4)18,
12(1)2,
12(1)3,
12(2)5,
12(2)7,
12(3)9,
12(4)17,
13(2)10
- test,
6(2)6,
7(2)5,
9(1)3,
9(3)10,
9(3)11,
11(4)13,
12(3)9,
13(1)4,
13(2)7,
13(3)13
- testing,
11(1)3,
11(4)18
- than,
5(2)146,
6(3)10,
7(3)8,
7(4)11,
7(4)13,
8(2)8,
8(4)16,
9(1)2,
9(2)7,
9(3)11,
9(3)12,
10(1)3,
10(1)4,
10(3)14,
11(2)4,
11(2)7,
11(3)8,
11(3)9,
11(4)13,
11(4)15,
12(2)7,
12(3)10,
12(4)16,
13(1)1
- time,
3(1)11,
3(1)66,
5(4)388,
6(4)3,
7(1)3,
7(3)9,
11(3)8,
12(1)4,
12(2)6,
12(3)12,
13(1)1,
13(2)6,
13(3)12
- to-English, Chinese-,
7(1)1,
8(1)4,
8(2)6
- training,
5(2)121,
6(3)11,
7(1)1,
8(1)3,
8(2)6,
8(2)7,
8(3)10,
9(2)5,
10(3)12,
10(3)13,
11(3)9,
12(1)1,
12(2)5,
12(3)9,
12(3)10,
13(1)2,
13(1)3,
13(1)4
- tuning,
10(1)3
- two,
5(2)89,
7(2)7,
7(3)8,
7(4)11,
7(4)12,
7(4)13,
8(1)4,
8(2)7,
8(4)17,
9(1)2,
9(3)11,
9(4)13,
10(1)2,
10(3)12,
10(3)14,
10(3)15,
10(4)20,
11(2)4,
11(2)5,
11(2)7,
11(3)8,
11(3)9,
11(3)11,
11(4)17,
12(1)1,
12(1)2,
12(1)4,
12(2)5,
12(3)10,
12(3)11,
12(4)16,
13(1)1,
13(1)3,
13(1)4,
13(2)6,
13(2)9,
13(3)11
- two-phase,
5(1)61,
12(1)3
- unstable,
7(3)10
- update,
11(2)5
- weight,
10(1)3,
10(2)9
- which,
5(2)89,
5(2)121,
6(2)6,
6(2)8,
6(3)10,
6(4)1,
7(1)3,
7(2)7,
7(3)10,
7(4)13,
8(1)2,
8(1)4,
8(2)9,
8(3)10,
8(3)12,
8(4)14,
8(4)16,
8(4)18,
9(1)1,
9(2)5,
9(3)12,
9(4)14,
9(4)15,
10(1)4,
10(1)6,
10(2)7,
10(2)8,
10(2)9,
10(3)12,
10(3)13,
10(3)14,
10(3)15,
10(4)17,
10(4)19,
10(4)20,
11(2)6,
11(3)8,
11(3)9,
11(4)13,
11(4)14,
11(4)15,
11(4)16,
11(4)18,
12(1)1,
12(1)3,
12(1)4,
12(2)5,
12(2)6,
12(2)7,
12(3)10,
12(3)11,
12(4)15,
12(4)17,
13(1)1,
13(2)8,
13(3)13,
13(4)18