Taku Fukushima

A web-based multilingual parallel corpus collection system for the medical field

We have developed a Web-based multilingual parallel corpus collection system named TackPad for medical workers and foreign patients. The number of foreign visitors and residents in Japan is increasing. However, it is not sufficient to provide linguistic support via medical translators to people who do not understand Japanese. In particular, poor communication may lead to medical errors. Moreover, there are limitations to a medical translators support. It is difficult to use machine translations in the medical field because of their inaccuracy. A system that uses parallel corpora with guaranteed accuracy appears to be useful. However, a huge amount of parallel corpora related to specialty fields and in several languages is required in the medical field. It would be difficult for a limited number of people to accumulate a sufficient amount of parallel corpora. Therefore, we have specially designed a support system named TackPad for the medical field; this system can accurately register parallel corpora and supply this data to other systems. TackPad adopts a Web 2.0 approach and can be used to tap peoples capabilities for knowledge accumulation. The purpose of TackPad is to collect parallel corpora; other systems can use the corpus via the Internet. We have obtained various responses from the result of the trial experiment. The experiment has demonstrated the practicality of the system, for example, people of various age groups and from various occupations can actually use the system.

Denoising autoencoder and environment adaptation for distant-talking speech recognition with asynchronous speech recording

In this paper, we propose a robust distant-talking speech recognition system with asynchronous speech recording. This is implemented by combining denoising autoencoder-based cepstral-domain dereverberation, automatic asynchronous speech (microphone or mobile terminal) selection and environment adaptation. Although applications using mobile terminals have attracted increasing attention, there are few studies that focus on distant-talking speech recognition with asynchronous mobile terminals. For the system proposed in this paper, after applying a denoising autoencoder in the cepstral domain of speech to suppress reverberation and performing Large Vocabulary Continuous Speech Recognition (LVCSR), we adopted automatic asynchronous mobile terminal selection and environment adaptation using speech segments from optimal mobile terminals. The proposed method was evaluated using a reverberant WSJCAMO corpus, which was emitted by a loudspeaker and recorded in a meeting room with multiple speakers by far-field multiple mobile terminals. By integrating a cepstral-domain denoising autoencoder and automatic mobile terminal selection with environment adaptation, the average Word Error Rate (WER) was reduced from 51.8% of the baseline system to 28.8%, i.e., the relative error reduction rate was 44.4% when using multi-condition acoustic models.

Development of multilingual interview-sheet composition system to support multilingual communication in medical field

Recently, the number of foreign residents and foreign visitors in Japan has been increasing every year. Consequently, the opportunities for communication amongst people whose native languages differ are increasing. In healthcare facilities, a paper-based multilingual interview sheet is used to facilitate communication between medical workers and foreign patients. However, this interview sheet has been found to be inadequate for such purposes. Moreover, Japanese medical workers find it difficult to understand the different languages written on a paper-based interview sheet. To resolve this problem, we have developed a multilingual interview-sheet composition system that uses parallel texts and machine translation. This system can convey essential patient information to a medical worker during consultation. The contributions of this study are as follows: (1) We have proposed a multilingual interview-sheet composition system that can be used for communication between medical workers and foreign patients. We have developed this system using both parallel texts and machine translation. (2) We showed that a patient is able to create a multilingual interview sheet using a parallel corpus and machine translation. (3) Because it may be difficult to indicate an affected area of the body using words alone, we suggest that affected areas be indicated by the user using a human-body image for more accurate communication.

Evacuation Support System for Everyday Use in the Aftermath of a Natural Disaster

Numerous natural disasters occur in Japan, such as earthquakes, typhoons, and volcanic eruptions. Information technology is expected to facilitate evacuation support when a disaster strikes, but networks cannot be used in many cases. Moreover, if a disaster occurs in a town that people do not frequent often, they may find it difficult to cope initially. It is difficult to access software online immediately at the time of a disaster. Thus, we have developed an evacuation support system called AkariMap, which can be used offline in the aftermath of a disaster. AkariMap is also a system that is suitable for everyday use. AkariMap has two functions: a function that notifies a user about evacuation support information each day and a disaster mode function that users can access at times of emergency. Based on the results of our experiments, we report the following findings. (1) Subjects increasingly accessed local evacuation support information for about 10 days. (2) Even if people did not use the notification function in the long-term, they were still accustomed to using the system. (3) If people need to find a shelter using AkariMap, they can identify a safer shelter by displaying the flooded areas.

Automatic Translation Repair Method for Improving Accuracy of Translated Sentences

In this study, we have developed an automatic translation repair method to automatically improve the accuracy of translations. Machine translation (MT) supports multilingual communication, however, it cannot achieve high accuracy. MT creates only one translated sentence, therefore, it is difficult to improve the accuracy of translated sentences. Our method creates multiple translations by adding personal pronouns to the source sentence and by using a word dictionary and a parallel corpus. In addition, it selects an accurate translation from among the multiple translations using the results of a Web search. As a result, the automatic translation repair method improved the accuracy of translated sentences, and its accuracy is greater than that of MT.

Proposal and Evaluation of an Extraction Method for Inaccurate Example Sentences Using a Web Search Engine for Multilingual Parallel Texts

In this study, we have proposed an extraction method for inaccurate example sentences using a Web search engine for multilingual parallel texts. We developed a multilingual parallel-text sharing system named Tack Pad for multilingual communication in the medical field. However, it should be noted that parallel texts created by people can be inaccurate. Hence, we cannot use these parallel texts in fields where high levels of accuracy are required. Moreover, it is difficult for people to evaluate the parallel texts enough because these are large in number. Therefore, we proposed and evaluated an extraction method for inaccurate example sentences. This method uses the contents on the Web as wisdom of crowds. It splits an example sentence into n-grams and uses the Web search engine to locate the split words. Moreover, this method uses two thresholds to detect several mistakes which are typographical errors, grammatical errors, and so on. The contributions of this paper are the following results: (1) We proposed an extraction method that improves the accuracy of the example sentences using the Web search engine and (2) We showed an improvement in the accuracy of the example sentences using two thresholds.