( )

Size: px

Start display at page:

Download "( )"

きみえつちかね
5 years ago
Views:

1 B4IM

2 ( )

3 (e.g., eat ) (e.g., arrest ),,, 10., B4IM2035, i

4 DCS Vector DCS Vector ii

5 1 Inoue DCS Tree DCS Tree DCS Tree DCS Tree iii

6 1 (PD) (CSPD, CSPD-X) iv

7 1 Web (1) it (?) banana (j) (1) A monkey (i) gets a banana (j), and eats it (?). it (?) banana (j) it (?) monkey (i) eat monkey (i) banana (j) (2) him (?) John (i) arrest 1

8 John (i) Bob (j) (2) John (i) killed Bob (j). Police arrested him (?). (2) kill kill arrest arrest John (i)

9 2 2.1 FrameNet[1] FrameNet FrameNet [2, 3] FrameNet [4, 5] 2.2 3

10 Peng Gigaword Corpus 1 [6] ILP Solver Inoue [7] 1 4

11 3 ( ) Inoue Tian DCS Vector 3.1 Inoue (, ) (,, ) (, ) (,, ) [8] Inoue (2) Inoue 1 (2) John (i) killed Bob (j). Police arrested him (?). (2) him (?) John (i) John kill Bob ( 1 g) Police arrest him him Bob John Bob John him ( 1 g ) 1 5

12 1: Inoue 3.2 DCS Vector Inoue (, ) (,, ) (3) there (?) restaurant (i) store (j) store (j) restaurant (i) restaurant (i) new Inoue (3) There is a new [restaurant (i) ] next to a grocery [store (j) ]. I will go there (?). new restaurant 6

13 DCS Vector[9] ( 2) (subj) (obj) DCS Vector DCS Tree DCS Tree ( 3) 2: 3: DCS Tree DCS Vector John killed Bob. DCS Tree 4 him 1 v john := (v bob M ARG M 1 COMP + v kill)m SUBJ M 1 ARG + v john (1) 1 v bob M ARG M 1 COMP Bob (v bob M ARG M 1 COMP + v kill)m SUBJ M 1 ARG Bob John Bob John 7

14 4: DCS Tree DCS Vector DCS Tree x 2 v x := v x + 1 n n i=1 n x y x M L, M 1 P v yi M Li M 1 P i (2) x, y DCS Tree x, y x x y y DCS Tree x, y 3 p(θ) = σ(v x M P M 1 L... M KM 1 N v y) (3) σ θ DCS Tree x, y v x, v y M P M 1 L... M KM 1 N NCE (Noise-contrastive estimation[10]) x, P 1, L 2,..., P 2i, L 2i+1, y P i, L i y 8

15 3.3 DCS Vector Tian x y y DCS Tree (2) DCS Vector police arrest him him Bob John Inoue him John killed Bob DCS Tree ( 5) Tian (2) John (i) killed Bob (j). Police arrested him (?). 5: DCS Tree DCS Tree Tian DCS Tree 3 9

16 4 4.1 ClueWeb 12 2 ClueWeb ClueWeb 12 Stanford CoreNLP [11] Stanford CoreNLP [12] 71.0% 87.0% DCS Tree DCS Tree

17 3. DCS Tree DCS Tree DCS Tree 5. (Type A) 6. Tian (Type B) Type A 14 Type B 7 1/10 Type A 106,154,102 Type B 61,998, *UNKNOWN* 26,

18 Pseudo-disambiguation (PD) (e.g., monkey ARG SUBJ eat COMP ARG banana watch ) Context-sensitive PD (CSPD) (e.g., arrest COMP ARG John John ARG SUBJ kill Bob Bob ARG COMP kill ) CSPD-X CSPD (e.g., police ARG SUBJ arrest COMP ARG John ARG SUBJ kill Bob ARG COMP kill ) Type B DCS Vector Type A /1024, 1/4096, 1/8, 1/512, orthogonal regularizer 1/16, inverse regularizer 1/16 4, DCS Tree DCS Tree 3 Type A Type B 12

19 2. DCS Tree DCS Tree DCS Vector 3. DCS Tree DCS Vector 4. DCS Tree DCS Vector 5. 2., 3. 2., 4. DCS Tree DCS Vector DCS Tree DCS Tree 0, 1, PD no matrix DCS Vector 4 random PD target context accuracy accuracy accuracy depth depth (no matrix) all : (PD)

20 1 target depth 1 target depth CSPD, CSPD-X 2 CSPD CSPD-X target context accuracy accuracy accuracy accuracy depth depth (no matrix) (no matrix) all all all all all all : (CSPD, CSPD-X) target depth context depth 1 no matrix depth 2 target depth, context depth all ClueWeb

21 DCS Tree DCS Tree DCS Tree DCS Tree 15

22 5 DCS Vector 16

23 17

24 [1] Collin F. Baker, Charles J. Fillmore, and John B. Lowe. The Berkeley FrameNet Project. ACL 98 Proceedings of the 36th Annual Meeting of the Association for Computational Linguistics and 17th International Conference on Computational Linguistics, Vol. 1, pp , [2] Cosmin Bejan and Sanda Harabagiu. Unsupervised Event Coreference Resolution with Rich Linguistic Features. Proceedings of the 48th Annual Meeting of the Association for Computational Linguistics, No. July, pp , [3] Altaf Rahman and Vincent Ng. Coreference Resolution with World Knowledge. Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies, Vol. 1, pp , [4] David Bean and Ellen Riloff. Unsupervised Learning of Contextual Role Knowledge for Coreference Resolution. Proceedings of the Human Language Technology Conference of the North American Chapter of the Association for Computational Linguistics (HLT-NAACL 2004), pp , [5] Mohit Bansal and Dan Klein. Coreference Semantics from Web Features. Proceedings of the 50th Annual Meeting of the Association for Computational Linguistics (ACL-2012), pp , [6] Haoruo Peng, Daniel Khashabi, and Dan Roth. Solving Hard Coreference Problems. Proceedings of the 2015 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Vol , pp , [7] Naoya Inoue, Ekaterina Ovchinnikova, Kentaro Inui, and Jerry Hobbs. Coreference Resolution with ILP-based Weighted Abduction. Proceedings of COL- ING 2012, Vol. 2, No. December 2012, pp ,

25 [8] Naoya Inoue, Yuichiro Matsubayashi, Masayuki Ono, Naoaki Okazaki, and Kentaro Inui. Modeling Context-sensitive Selectional Preference with Distributed Representations. In Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pp , [9] Ran Tian, Naoaki Okazaki, and Kentaro Inui. Learning Semantically and Additively Compositional Distributional Representations. Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics, Vol. 1, pp , [10] Michael U Gutmann and Aapo Hyvärinen. Noise-contrastive estimation of unnormalized statistical models, with applications to natural image statistics. Journal of Machine Learning Research, Vol. 13, No. Feb, pp , [11] Christopher D Manning, Mihai Surdeanu, John Bauer, Jenny Rose Finkel, Steven Bethard, and David McClosky. The stanford corenlp natural language processing toolkit. In ACL (System Demonstrations), pp , [12],,,,.. 22, pp ,

26 ,,.., December Kotaro Sakamoto, Madoka Ishioroshi, Hyogo Matsui, Takahisa Jin, Fuyuki Wada, Shu Nakayama, Hideyuki Shibuki, Tatsunori Mori and Noriko Kando. Forst: Question Answering System for Second-stage Examinations at NTCIR- 12 QA Lab-2 Task. Proceedings of the 12th NTCIR Conference on Evaluati on of Information Access Technologies, June 2016.,,,,,. 1 ( ). 22, March 2016.,,,,,. 2 ( ) 3 ( ). 22, March 2016.,,.. 21, March 2015.,,,. Project NEXT. 21, March

[1], B0TB2053, 20014 3 31. i

B0TB2053 20014 3 31 [1], B0TB2053, 20014 3 31. i 1 1 2 3 2.1........................ 3 2.2........................... 3 2.3............................. 4 2.3.1..................... 4 2.3.2....................