22 Wenjie Cao, Chengqing Zong and Bo Xu C-STAR: Consortium for Speech Translation Advanced Research IF: Interchange Format [12] C-STAR IF 1. IF IF [14

Transcription

1 Journal of Chinese Language and Computing, 14 (1) {caowj, cqzong, 1

2 22 Wenjie Cao, Chengqing Zong and Bo Xu C-STAR: Consortium for Speech Translation Advanced Research IF: Interchange Format [12] C-STAR IF 1. IF IF [14] (canned-text system) (template system) Kukich 1983 ANA (phrase-based system) McDonald 1980 MUMBLE Moore 1989 EES (feature-based system) Matthiessen 1983 PENMAN KPML Bateman Maier et. al., / 44% 22

3 [7] IF 2 IF C-STAR IF NESPOLE! IF speaker speech act concepts [17] IF J. Austin 1962 J. Searle 1969 [6] IF IF Fillmore IF IF IF head-mismatching [3] IF Underspecified Semantic Representation, USR IF IF IF IF IF IF IF arguments IF (1) c:request-information+departure+transportation (transportation-spec=(flight, identifiability=yes, destination=tokyo), time=(clock=(hours=2))) Does the flight to Tokyo leave at 2 o clock? 2 (2) a:greeting(greeting=hello) Hello.

4 24 Wenjie Cao, Chengqing Zong and Bo Xu IF (IF) 2. [15]

5 IF IF IF IF IF IF IF IF IF 1 IF IF IF IF IF IF

6 26 Wenjie Cao, Chengqing Zong and Bo Xu IF 3. P C A P Pattern IF C Constraints IF Concepts Arguments A Action IF P C IF IF P C A ((speaker = a ) ( speech act = give-information ) ( topic = availability + room )); //Pattern (exist(argument, room-spec' )); //Constraints ( (cat = clause) ( mood = declarative) ( tense = present) (voice = active) (process((type = possessive) (lex =#dic(have)) (args = (((case = pos) (pos=lex=(#get(argument, 'who') #dic(we))) // we (bel=lex =( #get (attribute, 'room-spec' )))),)), (!optional: pre_mod = ( time = #get ( attribute, 'time')))); [6]

7 IF Values *payment-methods* ((speechact=give-information),(topic=payment)); //Pattern (exist(method), value( method ) *payment-methods*); //Constraints ((cat=clause), (process.type=template), (lex= #dic(%pament) By #dic(%pament). )); [4] IF IF IF IF identifiability IF IF (type=[type], name=[name]); (exist_set=([arg_e_1], [arg_e_m],,[arg_e_n]), non-exist_set=([arg_n_1], [arg_n_2],,[arg_n_m])); ((cat=[phrase_type]), (c_rule=({left_cons=[sem_1]}, {right_cons=[sem_2]}),

8 28 Wenjie Cao, Chengqing Zong and Bo Xu (#dic([name], %head),( [arg_1], [arg_2],, [arg_i-1], *, [arg_i+1],, [arg_k])), ), (e_rule=({left_cons=[sem_1]}, {right_cons=[sem_2]}), (#dic([name], %head), ([Arg_1], [Arg_2],, [Arg_i-1], *, [Arg_i+1],, [Arg_k])),),); type argument concept argument concept name argument concept Constrains arguments [name] arguments [name] arguments Constrains c_rule e_rule cat np vp mp c_rule left_cons right_cons #dic([name], %head) [name] head * 3.2 ( ) ( ) 4.

9 3.2.1 SFG Halliday 1985 SFG Halliday participant Halliday 5 Halliday

10 30 Wenjie Cao, Chengqing Zong and Bo Xu IF IF 23 [16] process.type possessive mood declarative tense present voice active (possessor.type person.person first.number plural) (possessed.type object *room*.lex single.number single (.modifier.type describer *price-modifiers*.modifier.degree comparative.modifier.lex cheap )) We have a cheaper single room. mood tense voice

11 3.2.2 [6] the 4 C-STAR BETEC [12]

12 32 Wenjie Cao, Chengqing Zong and Bo Xu NESPOLE! BETEC IF Corpus Correct Understandable Wrong No result Accuracy HOTEL % TRAVEL % 1. 1 Understandable No result ( Correct + Understand able) / Total 1 5 IF 1 NESPOLE! IF C-STAR

13 AA [1] Lori Levin, et al., Balancing Expressiveness and Simplicity in an Interlingua for Task Based Dialogue, In Proceedings of Speech-to-Speech Translation Workshop at the 40th Annual Meeting of the Association of Computational Linguistics (ACL-02), Philadelphia, PA, July 2002, Pages [2] C-STAR Consortium, Dialogue Act Annotation, from htdocs/comp/, October 5, [3] Lori Levin, et al., The NESPOLE! Interchange Format, from nespole/ db/current/, Feb. 24th, [4] Hua Wu, Taiyi Huang, Chengqing Zong and Bo Xu, Chinese Generation in a Spoken Dialogue Translation System, Proceedings of COLING 2000, Pages [5] Emanuele Pianta and Lucia M. Tovena, XIG: Generating from Interchange Format using Mixed Representations, Proceedings of Euro Workshop on NLG, Toulous F, 1999, Pages [6],,, [7] Ehud Reiter, NLG vs. Templates, In Proc of the Fifth European Workshop on Natural-Language Generation (ENLGW-1995), Leiden, the Netherlands, Pages [8] E. Reiter and R. Dale, Building Applied Natural-Language Generation Systems, Journal of Natural-Language Engineering, March, 1997, Pages [9].., [10].., [11] [12] [13] L.M.Tovena and E.Pianta, Generating felicitous sentences from underspecified semantic representations, in Proceedings of the 3rd International Workshop on Computational Semantics, Tilburg NL, Genuary 1999, Pages

14 34 Wenjie Cao, Chengqing Zong and Bo Xu [14] John A. Bateman. Language Generation. In R. A. Cole, J. Mariani, H. Uszkoreit, A. Zaenen and V. Zue, eds, 'Survey of State of the Art in Human Language Technology', Cambridge University Press, chapter 4, Pages [15] Stephan Busemann and Helmut Horacek, A Flexible Shallow Approach to Text Generation. In: Eduard Hovy (ed.): Proceedings of the Nineth International Natural Language Generation Workshop (INLG '98), Niagara-on-the-Lake, Canada, August 1998, Pages [16] NESPOLE! T ravel Interchange Format Specification From specification.html, February, [17].,. Wenjie Cao, Chengqing Zong and Bo Xu National Laboratory of Pattern Recognition, Institute of Automation Chinese Academy of Sciences, Beijing , China {caowj, cqzong, xubo}@nlpr.ia.ac.cn Abstract: The Interlingua-based machine translation technique is an important approach to implement multi-lingua speech-to-speech (S2S) translation. The natural language generation (NLG) is one of the key components in the interlingua-based machine translation systems. This paper introduces our research on Chinese and English generation based on interlingua. In our system, we use the hybrid method of feature-based deep generation technique and template-based technique as our generation strategy, in which the deep generation technique ensures that the S2S translation system possesses the merits of flexibility and domain portability, and the template-based technique makes the system be more efficient. The experiment shows that the hybrid method is suitable for the natural language generation in the Interchange-Format (IF) based S2S translation system. Key words: Natural language generation, Micro-planning, Lexicalization, Surface realization