A Web Service for Evaluating the Level of Speech in Korean

Round 1

Reviewer 1 Report

This paper discusses a study on the system that shows the level of conversation compared with the level of other conversation. This system is based on the web and has the advangate of providing analysis results to users in real time. However, the following additional steps seem necessary.

The English expression of this paper should be more refined.

Difference between existing researcy and systems should be discussed in more detail. In particular, additional analysis of overseas research cases is required.

Author Response

 The English expression of this paper should be more refined.

    - We have studied the English expression of this paper and have corrected the problematic parts.

   2. Difference between existing researcy and systems should be discussed in more detail. In particular, additional analysis of overseas research cases is required.

    -  We have added an explanation of the characteristics of this study to the introduction. We have added domestic research as well as overseas cases.

Reviewer 2 Report

The paper proposed a Korean automatic speech analysis system based on natural language processing and web service. The automatic speech analysis service analyzes not only evaluate the overall language ability of the speaker but also the ability of individual domains such as sentence completion ability and vocabulary ability. Furthermore, according to the paper, the system provides a faster and more immediate service without sacrificing accuracy compared to human analysis.

(1)  According to section 3.2, there should have some common international evaluation measures to evaluate the level of the speech, which should be described in Induction section.

(2) In line 55, the authors mentioned different automated analysis systems, of which the specific technologies the systems used should be described more. It shouldn’t just say KCLA, KLA, what are the technologies behind the systems?

(3) In this paper, it frequently mentioned the natural language processing technology; while from line 170, it just means the NLP technologies are data pre-processing and morphological analyses. Are the morphological analyses enough for evaluating the level of speech? How about considering the semantics?

(4)  In section 2.2.2, each age division has 20 persons, is it enough?

(5)  It seems that the technology adopted is a kind of traditional machine learning, have the authors ever considered the deep learning technology?

(6) According to section 4.1, the format of the transcription is a little complex, if the authors provide a transcription template on the client page would be better.

(7) some grammatical errors should be revised.

Author Response

 According to section 3.2, there should have some common international evaluation measures to evaluate the level of the speech, which should be described in Induction section.

    - We added the reason why the measures were selected for the use of sentence evaluation in section 3.2

  2. In line 55, the authors mentioned different automated analysis systems, of which the specific technologies the systems used should be described more. It shouldn’t just say KCLA, KLA, what are the technologies behind the systems?

 - KCLA and KLA are little different from manual work of experts. We describe this in line 57-58.

3 In this paper, it frequently mentioned the natural language processing technology; while from line 170, it just means the NLP technologies are data pre-processing and morphological analyses. Are the morphological analyses enough for evaluating the level of speech? How about considering the semantics?

 - For sentence-level language analysis required by speech pathology, only morphological analysis is required up to now. Of course, for analysis beyond the sentence level, analysis of the topic of the conversation is necessary. This could be our future research topic.

4.  In section 2.2.2, each age division has 20 persons, is it enough?

 - Of course not. Now we keep collecting interview transcripts. However, because the interview should be leaded by experts, it is very expensive and time-consumiing.

5. It seems that the technology adopted is a kind of traditional machine learning, have the authors ever considered the deep learning technology?

 - We have applied this study to the RNN family in other researches and are preparing to publish the interim results in several conferences and journals.

6. According to section 4.1, the format of the transcription is a little complex, if the authors provide a transcription template on the client page would be better.

 - All user input are in Excel format, which is given to the interviewer with sufficient explanation. We will modify the format to make it easier for users.

7. some grammatical errors should be revised.

 - We read the article proofly to reflect your comments.

Thank you.