A Multitask Cross-Lingual Summary Method Based on ABO Mechanism
Round 1
Reviewer 1 Report
In the following, some concerns that I wish the authors can provide justifications.
1- What are the persistent problems with summary misordering and loss of key information in previous cross-lingual summarization research using a unified end-to-end model?
2- How does the proposed approach simplify multitasking in cross-lingual summarization?
3- What is the reinforced regularization method, and how does it improve the robustness of the model?
4- What is the targeted ABO mechanism, and how does it enhance the semantic relationship alignment and key information retention of cross-lingual summaries?
5- What evidence demonstrates the superior performance and ability of the proposed approach to improve cross-lingual sequencing on the professional domain dataset?
6- How does splicing monolingual and cross-lingual summary sequences as input help the model learn the core content of the corpus?
Acceptable English level.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 2 Report
The paper is sufficiently innovative but needs some improvement as follows to be ready for publication in the journal:
1-The introduction is too simple and could be improved with more strategies. Besides, the difference and contributions of research in comparison with the existing ones should be well clarified and highlighted.
2-There are several grammatical errors in this article and need to be revised to improve paper quality.
3- Why "targeted ABO mechanism " is hired? Please clarify its innovation compare to the prior works
4- Please explain and point out better about application of figure 2 in the text.
5-Please carefully clarify and give evidence about using equation 4 ?
6- How artificial intelligent as an innovative scheme can develop your study? It's extremely recommended to refer to the following paper and bring it into your introduction.
Neural Networks and Learning Algorithms in MATLAB
Modern Adaptive Fuzzy Control Systems
7- For, predicting the label distribution during training is crucial using equation 7 is very questionable. please clarify
8-Conclusions is very short. Contribution is not well emphasized. What is the main innovation of the work? Please point it out
9- What is your future direction
10- What is the limitation of your work? ( please point out and clarify all your examined dataset)
11-Figure 6 needs proof or valid reference. How and in which condition you used this data?
12- line 19 and 21 in Algorithm1 implementation are not clear at all and cause the discussible programming.
English includes mistakes. Please improve it.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
The main goal of the article is to improve cross-lingual summarization by simplifying the process and enhancing semantic relationship alignment and key information retention. The authors convert the translation task into equal proportion cross-lingual summary tasks, using both monolingual and cross-lingual summary sequences as input. They introduce a reinforced regularization method to increase the model's robustness and an Alignment Based Optimization (ABO) mechanism for better semantic alignment and information retention.
The article’s scientific soundness is good and the overall quality will be appropriate after some minor corrections in the paper. Please find some minor issues below.
“Reinforced Regularity Method” - maybe Reinforced Regularization Method ?
Please consider to introduce the ABO meaning, at its first usage in text e.g. Alignment Based Optimization (ABO) mechanism.
The phrase "which can achieve some degree of improvement" is a bit vague, it could be useful to specify what kind of improvement is being referred to, e.g. speed or accuracy.
“Transformer-based neural networks have been shown to share feature representations.”
Passive voice misuse.
Many times the space between subsequent sentences is missing.
The chosen evaluation and test data set size should be larger and/or k-fold cross validation should be used.
Please consider to save the figures in vector graphic format (e.g. eps).
Figure 6. is bad quality.
The content of the data set and the results of our model generation are shown in Figure 8. And we have published our dataset here:
https://github.com/leesin5079/Car-manual-CLS-dataset-1000
Inprecise statements.
Data Availability Statement: Data are contained within the article.
The data supporting the results are not contained within the article.
There is a URL in the paper to English language corpus and summary data files.
The overall quality of English language usage of the article is good. There are some minor mistakes and grammatical problems which can be easily corrected. Some or these issues are listed above among the suggestions.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
I thank the authors for their efforts in answering my concerns.
Reviewer 2 Report
Keep improvement .
Reviewer 3 Report
The authors addressed the reviewer comments. The paper quality issufficient for publication.