Genichiro Kikui

The ATR Multilingual Speech-to-Speech Translation System

In this paper, we describe the ATR multilingual speech-to-speech translation (S2ST) system, which is mainly focused on translation between English and Asian languages (Japanese and Chinese). There are three main modules of our S2ST system: large-vocabulary continuous speech recognition, machine text-to-text (T2T) translation, and text-to-speech synthesis. All of them are multilingual and are designed using state-of-the-art technologies developed at ATR. A corpus-based statistical machine learning framework forms the basis of our system design. We use a parallel multilingual database consisting of over 600 000 sentences that cover a broad range of travel-related conversations. Recent evaluation of the overall system showed that speech-to-speech translation quality is high, being at the level of a person having a Test of English for International Communication (TOEIC) score of 750 out of the perfect score of 990.

Subword-based Tagging by Conditional Random Fields for Chinese Word Segmentation

We proposed two approaches to improve Chinese word segmentation: a subword-based tagging and a confidence measure approach. We found the former achieved better performance than the existing character-based tagging, and the latter improved segmentation further by combining the former with a dictionary-based segmentation. In addition, the latter can be used to balance out-of-vocabulary rates and in-vocabulary rates. By these techniques we achieved higher F-scores in CITYU, PKU and MSR corpora than the best results from Sighan Bakeoff 2005.

A unified approach in speech-to-speech translation: integrating features of speech recognition and machine translation

Based upon a statistically trained speech translation system, in this study, we try to combine distinctive features derived from the two modules: speech recognition and statistical machine translation, in a loglinear model. The translation hypotheses are then rescored and translation performance is improved. The standard translation evaluation metrics, including BLEU, NIST, multiple reference word error rate and its position independent counterpart, were optimized to solve the weights of the features in the log-linear model. The experimental results have shown significant improvement over the baseline IBM model 4 in all automatic translation evaluation metrics. The largest was for BLEU, by 7.9% absolute.

Comparative study on corpora for speech translation

This paper investigates issues in preparing corpora for developing speech-to-speech translation (S2ST). It is impractical to create a broad-coverage parallel corpus only from dialog speech. An alternative approach is to have bilingual experts write conversational-style texts in the target domain, with translations. There is, however, a risk of losing fidelity to the actual utterances. This paper focuses on balancing a tradeoff between these two kinds of corpora through the analysis of two newly developed corpora in the travel domain: a bilingual parallel corpus with 420 K utterances and a collection of in-domain dialogs using actual S2ST systems. We found that the first corpus is effective for covering utterances in the second corpus if complimented with a small number of utterances taken from monolingual dialogs. We also found that characteristics of in-domain utterances become closer to those of the first corpus when more restrictive conditions and instructions to speakers are given. These results suggest the possibility of a bootstrap-style of development of corpora and S2ST systems, where an initial S2ST system is developed with parallel texts, and is then gradually improved with in-domain utterances collected by the system as restrictions are relaxed

Subword-Based Tagging for Confidence-Dependent Chinese Word Segmentation

We proposed a subword-based tagging for Chinese word segmentation to improve the existing character-based tagging. The subword-based tagging was implemented using the maximum entropy (MaxEnt) and the conditional random fields (CRF) methods. We found that the proposed subword-based tagging outperformed the character-based tagging in all comparative experiments. In addition, we proposed a confidence measure approach to combine the results of a dictionary-based and a subword-tagging-based segmentation. This approach can produce an ideal tradeoff between the in-vocaulary rate and out-of-vocabulary rate. Our techniques were evaluated using the test data from Sighan Bakeoff 2005. We achieved higher F-scores than the best results in three of the four corpora: PKU(0.951), CITYU(0.950) and MSR(0.971).

Japanese Dependency Parsing Using Sequential Labeling for Semi-spoken Language

The amount of documents directly published by end users is increasing along with the growth of Web 2.0. Such documents often contain spoken-style expressions, which are difficult to analyze using conventional parsers. This paper presents dependency parsing whose goal is to analyze Japanese semi-spoken expressions. One characteristic of our method is that it can parse self-dependent (independent) segments using sequential labeling.

Detecting Semantic Relations between Named Entities in Text Using Contextual Features

This paper proposes a supervised learning method for detecting a semantic relation between a given pair of named entities, which may be located in different sentences. The method employs newly introduced contextual features based on centering theory as well as conventional syntactic and word-based features. These features are organized as a tree structure and are fed into a boosting-based classification algorithm. Experimental results show the proposed method outperformed prior methods, and increased precision and recall by 4.4% and 6.7%.

A spoken language translation system: SL-trans2

1. I n t r o d u c t i o n An automatic telephone interpreting system wil l u n d o u b t e d l y be u s e d to o v e r c o m e c o m m u n i c a t i o n b a r r i e r s b e t w e e n people speaking different languages. Recently, great interest has been growing in this area [Saitoh88], [Waibel-91l, [Kitano-91], [Roe-92]. SLTRANS2 .1~ is an experimental system developed at ATR, which t rans la tes Japanese speech to English speech. It is composed of three basic components: speech recognition, translation and speech synthesis . This paper in t roduces the s y s t e m wi th e m p h a s i s on the t r a n s l a t i o n component. The discourse domain is a dialogue c o n c e r n i n g an i n t e r n a t i o n a l c o n f e r e n c e registrat ion. The d is t inc t ive f ea tu re s of the system are as follows. (1) Japanese continuous speech input can be recognized with h igh accuracy . Moreover , speaker independent recognition using speaker adaptation technique has been developed. (2) Var ious express ions pecul ia r to spoken language can be accepted and translated properly into the target language. In Japanese, the style of spoken sentences is general ly quite different from that of written texts. Spoken utterances are f ragmentary and include the speakers intention directly or indirectly. The system extracts the intent ion and then t ranscr ibes it to a proper expression in the target language. (3) Linguistic knowledge sources necessary for the translation are defined declaratively to the e x t e n t . Such d e f i n i t i o n i m p r o v e s h i g h modularity, readability and easy maintenance of knowledge description. In the next section, the system is overviewed and a brief description of the speech recognition

Feature structure based semantic head driven generation

This paper proposes a generation method for feature-structure-based unification grammars. As compared with fixed arity term notation. feature structure notation is more flexible for representing knowledge needed to generate idiomatic structures as well as general constructions. The method enables feature structure retrieval via multiple indices. The indexing mechanism, when used with a semantic head driven generation algorithm, attains efficient generation even when a large amount of generation knowledge must be considered. Our method can produce all possible structures in parallel, using structure sharing among ambiguous substructures.

Improving hmm-based extractive summarization for multi-domain contact center dialogues

This paper reports the improvements we made to our previously proposed hidden Markov model (HMM) based summarization method for multi-domain contact center dialogues. Since the method relied on Viterbi decoding for selecting utterances to include in a summary, it had the inability to control compression rates. We enhance our method by using the forward-backward algorithm together with integer linear programming (ILP) to enable the control of compression rates, realizing summaries that contain as many domain-related utterances and as many important words as possible within a predefined character length. Using call transcripts as input, we verify the effectiveness of our enhancement.