単語分散表現を用いた単語アライメントによる日英機械翻訳の自動評価尺度
7
0
0
全文
(2) Vol.2016-NL-229 No.20 2016/12/22. IPSJ SIG Technical Report. METEOR-Universal [3] METEOR-Universal WordNet. stemmer. PPDB [4]. METEOR BLEU Semantic Textual Similarity (STS). STS. 1:. Song. [10]. STS Song. STS. 3.2 One-hot. Song. 3. x. 3.1. One-hot. y. SV SV(a) =. 3.2 3.3. |a|. 1 ! ai |a|. (1). i=1. Song. a. [10]. x y. ai. SV. STSSV (x, y) =. 3.1 One-hot One-hot. One-hot. SV(x) · SV(y) |SV(x)||SV(y)|. 3.3 STS. 1. (2). Song. 0. 3.1. c 2016 Information Processing Society of Japan ⃝. 3.2. 2.
(3) Vol.2016-NL-229 No.20 2016/12/22. IPSJ SIG Technical Report. 3.3.3 Hungarian Alignment Similarity 3.3.1. 3.3.2. Whole Alignment Similarity. Maximum Alignment Similarity 3.3. 1 3.3. Hungarian Alignment Similarity (HAS) x. y. 2. 2 2. 1. 1.00. φ(xi , yj ). 2. 2 3.3 3.3.1 Whole Alignment Similarity. 2 Whole Align-. Hungarian. [7]. x. ment Similarity (WAS). xi. Hungarian. y. h(xi ). min(|x|, |y|) x. HAS(x, y). y |x||y|. |x|. ! 1 φ(xi , h(xi )) min(|x|, |y|). HAS(x, y) =. WAS(x, y). WAS(x, y) =. |x|. (6). i=1. 4.. |y|. 1 !! φ(xi , yj ) |x||y|. (3). 4.1. i=1 j=1. xi. yj. 4.2. φ(xi , yj ). 3. 3.3.2 Maximum Alignment Similarity 4.1. 3.3.1. WAT2015. NTCIR8 WAT2015. NTCIR8. –. 1. 600. ×200. 1,200. 3 ×100. 12. Maximum Alignment Similarity (MAS). Maximum Alignment. Similarity. x. y. y x 3.3. MAS(x, y). a. b. [8]. x. Google News 30. word2vec https://code.google.com/. y |a|. 1 ! MASasym (a, b) = max φ(ai , bj ) j |a|. archive/p/word2vec/ (4). i=1. MAS(x, y) =. 1 (MASasym (x, y) + MASasym (y, x)) (5) 2. BLEU METEOR NIST RIBES BLEU METEOR NIST RIBES. Asiya. [5]. NTT version 1.03.1. c 2016 Information Processing Society of Japan ⃝. 3.
(4) Vol.2016-NL-229 No.20 2016/12/22. IPSJ SIG Technical Report. 2:. 3: WAT2015. NTCIR8. http://www.kecl.ntt.co.jp/icl/lirg/ribes/. One-hot. index.html. 0.097. 4.2 2. 0.211. Whole Alignment Similarity. 0.332. Maximum Alignment Similarity. 0.235. Hungarian Alignment Similarity. 0.092. BLEU. 0.220. METEOR. 0.248. NIST. 0.204. RIBES. 0.261. 3 whole. Alignment Similarity maximum Similarity hungarian. Whole. Maximum Alignment. Hungarian Alignment Similarity. 1:. WAT2015. WAT2015. Whole Alignment Similarity 1. 2 One-hot. 1. 0.180 0.022. 2 Whole Alignment Similarity 3 3. Whole Alignment Similarity. 0.343. Maximum Alignment Similarity. 0.171. Hungarian Alignment Similarity. 0.075. BLEU. 0.225. METEOR. 0.211. 4 4. NIST. 0.150. RIBES. 0.368. 2: NTCIR8. 5. Whole Alignment Similarity (whole). 6. WAT2015 NTCIR8 WAT2015. 5.. NTCIR8 2. 3. c 2016 Information Processing Society of Japan ⃝. 1.00. Whole. Alignment Similarity 4.
(5) Vol.2016-NL-229 No.20 2016/12/22. IPSJ SIG Technical Report. One-hot. Whole Alignment Sim-. 0.983 0.979. Whole Alignment Similarity. 0.814. Maximum Alignment Similarity. 0.995. Hungarian Alignment Similarity. 0.656. BLEU. 0.531. METEOR. 0.973. NIST. 0.998. RIBES. 0.721. ilarity. BLEU. WAT2015 RIBES. RIBES WAT2015. NTCIR8 Whole Alignment Similarity Whole Alignment. Similarity 3. 3:. METEOR. 4. WAT2015. NTCIR8. WAT2015. Hungarian Alignment. Similarity One-hot. BLEU RIBES METEOR NIST Maximum. 0.543 0.645. Alignment Similarity. Whole Alignment Similarity. 0.515. RIBES. Maximum Alignment Similarity. 0.657. Hungarian Alignment Similarity. 0.097. BLEU. 0.075. METEOR. 0.482. NIST. 0.159. RIBES. 0.861. METEOR METEOR Whole Alignment Sim-. ilarity. Whole. 4: Alignment Similarity. NTCIR8. BLEU. RIBES BLEU. Whole Alignment Similarity. 0.30. Maximum Alignment Similarity Hungarian Alignment Similarity. 0.33 0.35. RIBES. 0.263 0.223. 5. 6. 0.068. 1. 5:. 2. WAT2015. WAT2015. Whole Alignment Similarity. 0.68. 7. 0.304. Maximum Alignment Similarity. 0.32. 0.152. Hungarian Alignment Similarity. 0.38. 0.073. WAT2015. 7 4. 1. 3. NTCIR8 [0,1]. 6: 1 NTCIR8. 2 2 BLEU. METEOR. RIBES RIBES NTCIR8. 3 3. 1. “that”. 2. c 2016 Information Processing Society of Japan ⃝. 5.
(6) Vol.2016-NL-229 No.20 2016/12/22. IPSJ SIG Technical Report. 1. : In both cases, the postoperative course was good. : In both the cases, postoperative course were good.. 2. : In the treatment, side effect was not recognized. : No side effect was noted during treatment.. 3. : It is found that the deformation is affected by the pair density distribution. : It was found that the deformation gave effects to the pairing density distribution.. 4. : The above configuration and operation, the water drops, even in the case where the window is not expected operation, it is possible to surely stopped, and the operation switch ( dn ) is operated and the window is opened, it is possible to a vehicle occupant moves out from possible. : With the above-described construction and operation, even when an automobile falls into water, the windows are surely stopped without performing unexpected operations, and can be surely opened by operating the operation switch ( dn ), thus enabling occupants to escape from the automobile. RIBES. METEOR. One-hot. WAS. MAS. 1. 0.83. 0.92. 0.90. 0.89. 0.94. 1.00. 2. 0.23. 0.37. 0.55. 0.73. 0.71. 1.00. 3. 0.30. 0.37. 0.64. 0.69. 0.78. 0.20. 4. 0.75. 0.25. 0.56. 0.11. 0.61. 0.80. 7: RIBES METEOR. 4. Whole Alignment Similarity 4. Whole Alignment. Similarity. Whole Alignment. Similarity [1]. RIBES. One-hot 3. NTCIR8. NTCIR8. [2]. NTCIR8 [3]. 6. [4]. [5]. Whole Alignment Similarity. c 2016 Information Processing Society of Japan ⃝. [6]. Satanjeev Banerjee and Alon Lavie. METEOR: An Automatic Metric for MT Evaluation with Improved Correlation with Human Judgments. In Proceedings of the ACL Workshop on Intrinsic and Extrinsic Evaluation Measures for Machine Translation and/or Summarization, pp. 65–72, 2005. Chris Callison-Burch, Miles Osborne, and Philipp Koehn. Re-evaluating the Role of BLEU in Machine Translation Research. In Proceedings of the 11th Conference of the European Chapter of the Association for Computational Linguistics, pp. 249–256, 2006. Michael Denkowski and Alon Lavie. Meteor Universal: Language Specific Translation Evaluation for Any Target Language. In Proceedings of the Ninth Workshop on Statistical Machine Translation, pp. 376–380, 2014. Juri Ganitkevitch, Benjamin Van Durme, and Chris Callison-Burch. PPDB: The Paraphrase Database. In Proceedings of the 15th Annual Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pp. 758–764, 2013. Jesus Gimenez and Lluis Marquez. Asiya: An Open Toolkit for Automatic Machine Translation (Meta)Evaluation. In The Prague Bulletin of Mathematical Linguistics, pp. 77–86, 2010. Hideki Isozaki, Tsutomu Hirao, Kevin Duh, Katsuhito Sudoh, and Hajime Tsukada. Automatic Evaluation of Translation Quality for Distant Language Pairs. In Proceedings of the 2010 Conference on Empirical Methods in Natural Language Processing, pp. 994–952, 2010.. 6.
(7) IPSJ SIG Technical Report. [7]. [8]. [9]. [10]. Vol.2016-NL-229 No.20 2016/12/22. Harold W. Kuhn. The Hungarian Method for the assignment problem. In Naval Research Logistics Quarterly, pp. 83–97, 1955. Tomas Mikolov, Kai Chen, Greg Corrado, and Jeffrey Dean. Efficient Estimation of Word Representations in Vector Space. In Proceedings of Workshop at ICLR, 2013. Kishore Papineni, Salim Roukos, Todd Ward, and WeiJing Zhu. BLEU: a Method for Automatic Evaluation of Machine Translation. In Proceedings of the 40th annual meeting on association for computational linguistics. Association for Computational Linguistics, pp. 311–318, 2002. Yangqui Song and Dan Roth. Unsupervised Sparse Vector Densification for Short Text Similarity. In Proceedings of the 2015 Annual Conference of the North American Chapter of the ACL, pp. 1275–1280, 2015.. c 2016 Information Processing Society of Japan ⃝. 7.
(8)
関連したドキュメント
(Construction of the strand of in- variants through enlargements (modifications ) of an idealistic filtration, and without using restriction to a hypersurface of maximal contact.) At
It is suggested by our method that most of the quadratic algebras for all St¨ ackel equivalence classes of 3D second order quantum superintegrable systems on conformally flat
Answering a question of de la Harpe and Bridson in the Kourovka Notebook, we build the explicit embeddings of the additive group of rational numbers Q in a finitely generated group
Next, we prove bounds for the dimensions of p-adic MLV-spaces in Section 3, assuming results in Section 4, and make a conjecture about a special element in the motivic Galois group
Transirico, “Second order elliptic equations in weighted Sobolev spaces on unbounded domains,” Rendiconti della Accademia Nazionale delle Scienze detta dei XL.. Memorie di
Then it follows immediately from a suitable version of “Hensel’s Lemma” [cf., e.g., the argument of [4], Lemma 2.1] that S may be obtained, as the notation suggests, as the m A
In our previous paper [Ban1], we explicitly calculated the p-adic polylogarithm sheaf on the projective line minus three points, and calculated its specializa- tions to the d-th
Our method of proof can also be used to recover the rational homotopy of L K(2) S 0 as well as the chromatic splitting conjecture at primes p > 3 [16]; we only need to use the
