A comparative study of machine translated subtitles based on the user-centered approach: a case study between Bilibili and YouTube

doi:10.21203/rs.3.rs-2179598/v1

Bilibili has become a famous video-sharing platform in China, attracting a growing number of visitors. Expect for the videos made by uploaders themselves, an increasing number of videos in different languages are uploaded. This year, the platform announced that automatic translation can be used for free to offer an alternative solution to the language barrier. Previous research has been done to test the impact machine-translated subtitles of TED (Technology, Entertainment, and Design) or MOOCs (Massive Open Online Courses) have on the users’ reception. Very little attention has been paid to do a comparative study between different platforms. This research did a pilot study to compare machine translated subtitles on Bilibili (MTS-B) with those on YouTube (MTS-Y). Based on the human evaluation of the quality and the questionnaire conducted online, the author analyzed the relationship between the quality, users’ comprehension and attitude towards machine translated subtitles. The results show that the quality has little impact on the users’ comprehension on the video. However, when MTS-Y performs better in accuracy, participants had a better attitude towards it.

Machine translation

subtitles

reception

quality

user-centered approach

Bilibili, nicknamed B site, has become a major video-sharing website in China, which hosts videos on various themes, including movies, music, technology, science, anime. According to websites ranking on Similarweb¹, bilibili.com is ranked number 2 on the list in China while baidu.com is number 1. According to Bilibili’s Investor Presentation (Q1, 2022)², the number of average monthly active users reached 293.6 million. The number of videos uploaded every month is about 10 million. Besides, a great number of uploaders repost videos in various languages and a growing number of foreigners come to Bilibili. Therefore, several projects have addressed audiovisual translation on Bilibili.com in the last decade. For instance, Yang (2020) has analyzed danmaku subtitling on Bilibili. She describes danmaku subtitling as a new form of online collaborative translation, except for fansubbing and professional subtitling. Based on danmaku subtitling research, Yang (2021) carried out a case study of a BBC documentary series to observe the characteristics of danmaku subtitling, which has been accepted by the audience as a timely language assistance. Ding et al. (2021) have investigated the user-generated translation activity on Bilibili via questionnaires and follow-up interviews. Their study shows that participants, also Bilibili content creators (uploaders), break down linguistic borders for viewers and provide them with different perspective of the content. However, for users, it takes time to wait for the danmaku subtitling, fansubbing and professional subtitling. In this case, Machine Translation (MT) could be a solution. On 15 July 2022, Bilibili announced that automatic translation could be used for some of the newly uploaded videos, mainly translating English, Japanese and Korean into Simplified Chinese, receiving a considerable amount of positive comments.³

More and more audience on Bilibili come to know and use this newly language assistance. Therefore, there is a need to analyze machine translation on Bilibili. This research will do a comparative study between MT subtitles on Bilibili and YouTube based on the user-centered approach, using human quality evaluation and survey.

¹Similarweb: https://www.similarweb.com/website/bilibili.com/#overview

² Bilibili 1Q22 Investor Presentation. (27 August 2022). Retrieved from http://ir.bilibili.com/financial-information/quarterly-results.

This section gives a brief introduction of the user-centered approach and machine translation for subtitling to provide the background of the research.

2.1 User-centred approach

The idea of taking users into account has been used and developed in research for many years. Based on UTAUT (Unified Theory of Acceptance and User of Technology) and APEC (aesthetic, practical, emotional, and cognitive) theories, Liu et al. (2021) take Bilibili as a case study to investigate the influencing factors of the user experience. The results of a questionnaire survey show that the relationship between content and user experience and the one between function and user experience is affected by the user’s emotion, while the influence of website interaction design and environment are not significant. Other scholars have expanded the user-centered approach in translation studies. Through eye tracking and the post-task questionnaire, Doherty and O’Brien (2012) conclude that there is a difference in usability between the source text and raw machine translated text for three measures: goal completion, user satisfaction and efficiency. Later, Suojanen et al. (2015) propose user-centered translation (UCT), including usability testing and reception research. Some practical tools and methods are offered in this book, such as eye tracking, questionnaires and interviews, for translators to consider end users throughout the translation process. They also point out that AVT provides an example of the relation between user experience and reception studies. Inspired by the UCT approach, Castilho (2016) investigates the acceptability evaluation of MT, including usability, quality and satisfaction. The results show that the raw machine translated versions are usable and the light post-editing positively affected acceptability for Simplified Chinese. Although more and more scholars take end users into account as the key factors in machine translation studies (e.g., Sánchez-Gijón and Kenny 2022, Ramírez-Sánchez 2022), the user-centered approach has attracted little attention in machine translation for subtitling.

2.2 Subtitling and Machine Translation

In the 21^st century, audiovisual translation (AVT) has become a fundamental necessity. Of the several modes available in AVT, subtitling is the most commonly used probably. It has been recognized and acknowledged by scholars to do research. Díaz Cintas (2010) states new trends in subtitling that any technical advances will probably influence the subtitling process, such as developments in voice and speech recognition. Automated translation in subtitling may be a promising field as a wider scope of genres are distributed audiovisually. Recent years have witnessed a change in the subtitling industry. The rising demand for multilingual content and freely available MT has raised the visibility of MT, especially since the automated translation of subtitles could be used from 2008, because of Google Translate’s integration into YouTube⁴. However, Burchardt et al. (2016) discovers that MT is used more widely in text translation than in audiovisual translation. They discussed challenges of using MT in AVT, such as domain and genre, lack of visual context, oral style and lack of context. Although MT output quality is affected by the grammatical irregularity in audiovisual texts and the language pair, Bywood et al. (2017) stress that MT systems are useful to increase productivity. Nurminen and Koponen (2020) also point out that the use of MT may promote linguistic accessibility of audiovisual media.

Nevertheless, more and more researchers turn to do theoretical and empirical research in machine translation for subtitling. Burchardt et al. (2016) have provided an overview of methods to evaluate MT and introduced the Multidimensional Quality Metrics (MQM) for evaluating mechine translated subtitles. The MQM Issue Types consist of accuracy, design, fluency, style and timing. Karakanta (2022) reviews previous experimental approaches in MT for subtitling, suggesting that all dimensions of subtitling should be considered, including content-related parameters, presentation and time parameters. Other Scholars start to do practical research for evaluating MT output in subtitling. According to Hiraoka and Yamada (2019), the quality of raw MT output for TED (Technology, Entertainment, and Design) subtitling, from Japanese to English, is “acceptable” on average, using human evaluation and BLEU (Bilingual Evaluation Understudy). It means that adequate information with audiovisual elements is provided. Hu et al. (2020) have done research on machine translated subtitles of MOOCs (Massive Open Online Courses). Through the combination of questionnaires and eye-tracking, they investigated the perception and reception of subtitled content. From their result, there was no significant difference in reception metrics among raw MT subtitles, PEMT (fully post-edited machine-translated) subtitles and HT (human-translated) subtitles. Besides, most participants held a positive attitude towards the MT subtitles. Previous research demonstrates the need to do empirical study in machine translation for subtitling. However, most of them focused on one platform (see Hiraoka and Yamada 2019, Hu et al. 2020). Thus, in this study, a comparative study between machine translated subtitling on different platforms will be done. According to the newest most visited Arts and

Entertainment websites ranking analysis on Similarweb⁵, Bilibili is 3rd on the list in the world while YouTube is ranked number 1. For audience demographics, there are some similarities between bilibili.com and youtube.com, such as gender distribution and age distribution (see Figure 1). The statistics indicate that a comparative study of Bilibili and YouTube is suitable for the research.

Therefore, this paper attempts to do a comparative study of machine translated subtitles between Bilibili and YouTube, based on the user-centered approach.

³ Bilibili. (15 July 2022). The function of automatic translation is available now! https://www.bilibili.com/read/cv17581547?from=search&spm_id_from=333.337.0.0

⁴ YouTube UK Blog. (1 November 2008). Auto Translate Now Available For Videos With Captions. http://youtubeukblog.blogspot.co.uk/2008/11/auto-translate-now-available-for-videos.html

In this research, each aspect will be explained in the following part. Accuracy refers to contextual errors of the machine translated subtitles, in terms of addition, mistranslation and omission. Design refers to the segment length, which means the number of characters in each line. Fluency consists of grammar (such as morphology, part of speech, agreement, word order and function words), spelling (such as capitalization) and typography. Style refers to the specific style used in target language, such as register and collocations. Timing refers to the characters per second (CPS). The source text and the subtitle generated by CC subtitling group are used as the reference for the translator to evaluate the machine translated subtitles. All these aspects will be evaluated by the author, who is an experienced translator.

3.1.2 Reception of the audience

The experiment is conducted online so only questionnaire will be used to evaluate the audience’s reception. The questionnaire in the study is adapted from Hu et al. (2020) and designed by the author, including comprehension testing and attitude questions. According to Hu et al. (2020), comprehension testing and attitude questions can be used to investigate some part of reaction and repercussion of reception. In this study, comprehension refers to what is understood by the viewer and attitude refers to viewers’ attitudes and beliefs towards the subtitles they are received.

3.2 Research setting and Hypothesis

Followed by 13.9 million users, Crash Course is a well-known channel on YouTube. By presenting information in a fast-paced format, it transforms the traditional textbook model. On Bilibili, a great number of uploaders repost videos of Crash Course and some of them translate the subtitles for free to help Chinese learners. Therefore, their videos are suitable for a comparative study. The video selected for this research was entitled ‘Crash Course World Mythology Preview’, which introduces the key point of world mythology on Crash Course channel. It lasted just under two minutes (1’58) and was composed of colourful images, which is suitable for the research. It was translated automatically from English into Simplified Chinese on Bilibili and YouTube using the automatic function.

The whole research involved two experiments. In quality evaluation experiment, the machine translated subtitles

on Bilibili (MTS-B) and YouTube (MTS-Y) were downloaded by the author and analyzed according to MQM. In reception experiment, the author recorded the videos with machine translated subtitles on Bilibili and YouTube, then uploaded them on Baidu Netdisk so that the participants were not aware of which of the subtitles is from Bilibili or YouTube. The questionnaire was posted on Bilibili to recruit Chinese native speakers as volunteers. The availability sample consisted of 44 volunteers. They were randomly assigned to the machine translated subtitles on Bilibili (namely Group MTS-B) and machine translated subtitles on YouTube (namely Group MTS-Y) to complete the questionnaire. The experiment involved two steps. In Step 1, the participants completed an online English proficiency test⁶ to measure English proficiency. In Step 2, participants were asked to provide their information, then watched the video on Baidu Netdisk and completed an online questionnaire, conducted through 问卷星 (www.wjx.cn) and administered in Chinese. To get objective results, all responses were anonymous.

The questionnaire consisted of three parts: demographic information, comprehension testing and attitude survey. Part one is designed to collect demographic information including gender, age, educational level, English proficiency and their experience of translating subtitles. It aims to find whether there is relationship between demographic data, their comprehension and attitude. Part two is inspired by Hu et al. (2020), including some comprehension questions to investigate the participants understanding about the machine translated content on Bilibili and YouTube. Part three is designed to investigate the attitude of the participants towards different machine translated subtitles.

Based on the reseach setting, several hypotheses have been derived as follows:

1. The machine translated subtitles on YouTube have better quality than the machine translated subtitle on Bilibili according to MQM.

2. If the machine translated subtitles on one platform perform better in quality, the level of comprehension is higher.

3. If the subtitle performs better in accuracy, the audience will have a better attitude towards it.

⁵Similarweb: https://www.similarweb.com/website/bilibili.com/vs/youtube.com/#overview

4.1. Quality evaluation

4.1.1 Accuracy and Fluency

In this part, the source text and the subtitle generated by CC subtitling group (CCS) are used as the reference for the quality evaluation. Comparing MTS-B, MTS-Y with the source text the CCS, the translator calculated the number of lines with errors in terms of accuracy (addition, mistranslation and omission) and fluency (grammar and spelling), which was shown in Table 1. There are 47 lines in MTS-B while 46 lines in MTS-Y. For example, if there are grammar mistakes in line 1, the number of lines in Grammar will be 1. If all lines in MTS-B have mistranslation, the number in Mistranslation will be 47. Table 1 shows that both MTS-B and MTS-Y have no spelling. There are less errors in terms of accuracy and fluency in MTS-Y than in MTS-B.

Table 1. The number of lines with errors

	MTS-B	MTS-Y
Accuracy
Addition	0	1
Mistranslation	12	9
Omission	2	1
Fluency
Grammar	4	3
Spelling	0	0
Total	18	13

For typography, the author compared MTS-B with MTS-Y mainly in the number of lines and position on screen, punctuation usage, and subtitle type. Figure 3 shows some examples of MTS-B and MTS-Y. For the number of lines, there is only one line per subtitle in MTS-B while two lines per subtitle in MTS-Y. Besides, both MTS-B and MTS-Y are placed at the bottom of the screen. However, subtitles are placed at the center on Bilibili while on YouTube, subtitles are on the left. MTS-B seldom use punctuation, only including three commas and one question marks. However, MTS-Y uses much more punctuation, including 27 commas, which seems less natural. Furthermore, MTS-B are edited subtitles while MTS-Y are verbatim subtitles. All these aspects show that MTS-B is cleaner and simpler than MTS-Y.

4.1.2 Design, Timing and Style

The segment length (the number of characters per line) and the characters per second (CPS) were calculated by the audiovisual translation software-ysj(人人译视界)⁷. In this study, a one-way ANOVA (Analysis of Variance) was carried out to test the difference of the segment length and the CPS between MTS-B and MTS-Y. In this case, the condition is set that α= 0.05 and the results were calculated by Excel (see Table 2).

Table 2 The segment length and CPS of MTS-B and MTS-Y

	Max	Min	Mean	Variance	P-value
Length
MTS-B	24	1	9.79	34.91	0.28
MTS-Y	30	5	10.93	17.53	0.28
CPS

MTS-B

8

1

4.11

2.53

0.24

MTS-Y

14

1

4.57

4.43

Although there is significant difference of variance between the segment length of MTS-B and MTS-Y, the p-value is much higher than 0.05, which shows no significant difference between the segment length of MTS-B and MTS-Y. Furthermore, there is also no significant difference between the CPS of MTS-B and MTS-Y.

When it comes to style, this study mainly analyzes the register. The video is talking about the introduction of world mythology crash course and the target readers are all online users who are interested in this course. Although the course is talking about history, the speaker uses personnal pronouns, such as ‘I’, ‘we’, and ‘you’and hopes to get closer to the audience. Therefore, the register of the video is semi-formal. Both subtitles on Bilibili and YouTube are automatically translated by machine and most of significant difference between MTS-B and MTS-Y.

To conclude, MTS-Y performs better in accuracy and the grammar and spelling of fluency. However, MTS-B shows cleaner and simpler subtitles than MTS-Y. There is no significant difference between MTS-B and MTS-Y, in terms of design, timing and style. Therefore, hypothesis 1 is not all correct. All the aspects mentioned above will be considered in the following part to find out the relationship between quality and the reception of the audience.

4.2 Demographic profile

The study collects responses from Group MTS-B and Group MTS-Y. There are 22 participants in each group. A total of 44 questionnaires were completed and valid, so the response rate is 100%. Table 3 shows the demographics of two groups, including gender, age, English proficiency, the education level and the experience of translating subtitles.

Table 3 Demographics of participants in Group MTS-B and Group MTS-Y

	Group MTS-B (n=22)		Group MTS-Y (n=22)
Gender
Male	4	18.18%	8	36.36%
Female	18	81.82%	14	63.64%
Total	22	100%	22	100%
Age
18-24	8	36.36%	11	50%
25-34	14	63.64%	8	36.36%
35-44	0	0	3	13.64%
Total	22	100%	22	100%
English Proficiency
Level A	8	36.36%	8	36.36%
Level B	9	40.91%	9	40.91%
Level C	5	22.73%	5	22.73%
Total	22	100%	22	100%
Highest education degree earned
High School Diploma	1	4.54%	0	0
Bachelor’s or College Degree	14	63.64%	14	63.64%
Master’s Degree	7	31.82%	8	36.36%
Total	22	100%	22	100%
Experience of Translating Subtitles
No	19	86.36%	19	86.36%
Yes	3	13.64%	3	13.64%
Total	22	100%	22	100%

In this study, it is surprising that the information of English proficiency and the experience of translating subtitles in two groups are exactly the same, although the participants are assigned randomly. For two groups, most of the participants don’t have any experience of translating subtitles, which shows that they can give comments as the general audience instead of professions. There are other similarities between two groups. For example, males are much less than females and the majority of the participants got Bachelor or College Degree. However, participants aged 18-24 in Group MTS-B are less than those in Group MTS-Y, while participants aged 25-34 in Group MTS-B are more than those in Group MTS-Y.

4.3 Comprehension testing

The video is just the review of crash course world mythology (1’58), so there are only two questions in this part. The comprehension questions aim to test the participants’ understanding of the key points of the video, which contain conjunctions. Here are these two questions:

6. This crash course is going to address which of the following questions:

A. where children come from

B. how we got death and disease

C. what the deal is with unicorns and virgins

D. all of the above

7. What is the key point of this crash course?

A. tick all of the mythological locales off of the geographical list one by one

B. see the similarities in the stories that different people tell about the universe

The results are shown on Figure 4. In question 6, D is the correct answer and presents a parallel relation in the video. In question 7, B is the correct answer and stressed in the video using ‘though’. For the comprehension questions, Group MTS-B shows the similarity in comprehension testing with Group MTS-Y. For the quality mentioned above, MTS-Y performs better than MTS-B in terms of accuracy and grammar and spelling of fluency. For typography, MTS-B shows cleaner and simpler subtitles than MTS-Y. However, it seems that the accuracy and the fluency of the subtitles have little impact on the comprehension of the audience, which goes against the hypothesis 2. Moreover, demographics of two groups show differences in gender and age, but it seems that they don’t affect the comprehension of the audience.

This part consists of 7 five-point Likert scale statements as well as one open question. Table 4 shows the distribution of different answers from Likert scale statements for Group MTS-B and Group MTS-Y in percentage form (absolute values are also shown in parentheses). For instance, if all the participants in Group MTS-B chose ‘Strongly agree’ for all the 7 statements, then this scale would have been chosen 154 times (22 ×7). However, it was chosen 42 times, less than one third, by Group MTS-B.

Table 4. Attitude survey results from Group MTS-B and Group MTS-Y

	Strongly agree	Agree	Neutral	Disagree	Strongly disagree
Group MTS-B	27.27% (42)	39.61% (61)	28.57% (44)	4.55% (7)	0
Group MTS-Y	34.42% (53)	43.51% (67)	18.18% (28)	3.89% (6)	0

According to Table 4, for Group MTS-B and Group MTS-Y, over half of the participants chose ‘Strongly agree’ and ‘agree’, while no on chose ‘Strongly disagree’, indicating that participants’ attitudes to MTS-B and MTS-Y are mostly positive. In respect of the options ‘Strongly agree’ and ‘agree’, Group MTS-Y has higher number than Group MTS-B. It is surprising that in Group MTS-B, there are more participants who chose ‘neutral’ than those who chose ‘strongly agree’.

For each statement, the percentage of agreements per group was calculated. In Group MTS-B, nearly half of the participants chose ‘neutral’ for the statement 8, much more than those in Group MTS-Y who chose ‘neutral’, indicating that MTS-B are less helpful for participants to fully understand the contents of the crash course.

For the open question, each participant per group was asked to give comments to the subtitles they are received. Although participants in MTS-B express less agreements than those in MTS-Y, 7 of them showed no comments towards MTS-B, more than those participants in MTS-Y. Other participants show their opinions towards the video they are received. For MTS-B, 3 participants noticed that there are some errors in accuracy, while 3 participants advise that it is necessary to provide bilingual subtitle. Some of them noticed the grammar mistakes while others cared about the typography of the font. For MTS-Y, most participants cared about typography of the subtitle, for example, the position and verbatim subtitles are not comfortable for reading. 2 of them noticed that grammar mistakes and suggest the bilingual subtitle. However, none of the participants mentioned the accuracy.

It can be seen that participants had a better attitude towards MTS-Y than MTS-B and they don’t have comments on the accuracy of MTS-Y. From the previous results on quality evaluation, MTS-Y performs better in accuracy than MTS-B. Therefore, the result is consistent with hypothesis 3 that if the subtitle performs better in accuracy, the audience will have a better attitude towards it.

Although most participants said that the verbatim subtitles as well as their position in the screen are not comfortable, they show not significant differences in comprehension testing.

⁶Website of Cambridge Assessment English: https://www.cambridgeenglish.org/test-your-english/general-english/

⁷ysj software: achieved from https://www.1sj.tv/html/otherService/download.html?version=1665739518000

This research has done a pilot study comparing machine translated subtitles on Bilibili with those on YouTube. It tests 3 hypotheses in total and investigates the results on the quality evaluation and the reception of the audience. For the hypothesis 1, the results show that MTS-Y performs better in accuracy and the grammar and spelling of fluency. However, MTS-B shows cleaner and simpler subtitles than MTS-Y. There is no significant difference between MTS-B and MTS-Y, in terms of design, timing and style. For the hypothesis 2, the results show that the accuracy and the fluency of the subtitles on Bilibili and on YouTube have little impact on the comprehension of the audience. For the hypothesis 3, it can be seen that when MTS-Y performs better in accuracy, the audience had a better attitude towards it.

On the whole, most participants could get key points and understand the contents of the video through the machine translated subtitles. Although the automated translation was put into use on Bilibili this year, the quality has little impact on the users’ comprehension on the video. However, when MTS-Y performs better in accuracy, participants had a better attitude towards it.

However, this research is a pilot study and it can be expanded in many ways. For instance, the sample of the experiment can be larger and more participants should be recruited. For the reception aspect, eye tracking can be used to collect more quantitative data. Moreover, more language pairs can be investigated because automatic translation on the video platform Bilibili is developing and improving, including Japanese into Chinese and Korean into Chinese. Lastly, bilingual subtitles could be another aspect for reception study.

No potential conflict of interest was reported by the author.

Acknowledgments

The author wishes to thank all the participants who took part in the experiment, for their time and feedback.

Competing Interests Statement

The author declares that she has no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Funding declaration

This work received no specific grant from any funding agency, commercial or not-for-profit sectors.

Burchardt, A., Lommel, A., Bywood, L., Harris, K., & Popović, M. (2016). Machine translation quality in an audiovisual context. Target: International Journal of Translation Studies, 28(2), 206–221.
Bywood, L., Georgakopoulou, P., & Etchegoyhen, T. (2017). Embracing the threat: machine translation as a solution for subtitling. Perspectives, 25(3), 492–508.
Castilho, S. (2016). Measuring acceptability of machine translated enterprise content [Doctoral dissertation, Dublin City University].
Díaz Cintas, J. (2010). Subtitling. In Y. Gambier, & Van L. Doorslaer (Eds.), Handbook of translation studies (pp. 344–349). John Benjamins.
Díaz Cintas, J., & Remael, A. (2020). Subtitling concepts and practices. Routledge.
Ding, N., Yang, Z., Li, S., & Zhang, A. (2021). Where translation impacts: The nonprofessional community on Chinese online social media–A descriptive case study on the user-generated translation activity of Bilibili content creators. Global Media and China, 6(2), 171–190.
Doherty, S., & O'Brien, S. (2012). A user-based usability assessment of raw machine translated technical instructions. Proceedings of the 10th Conference of the Association for Machine Translation in the Americas: Commercial MT User Program. Association for Machine Translation in the Americas.
Hiraoka, Y., & Yamada, M. (2019). Pre-editing plus neural machine translation for subtitling: effective pre-editing rules for subtitling of TED Talks. In M. Forcada, A. Way, J. Tinsley, D. Shterionov, C. Rico, & F. Gaspari (Eds.), Proceedings of Machine Translation Summit XVII: Translator, Project and User Tracks (pp. 64–72). European Association for Machine Translation.
Hu, K., O’Brien, S., & Kenny, D. (2020). A reception study of machine translated subtitles for MOOCs. Perspectives, 28(4), 521–538.
Karakanta, A. (2022). Experimental research in automatic subtitling: At the crossroads between machine translation and audiovisual translation. Translation Spaces, 1, 1–24.
Liu, W., He, Z., & Liu, M. (2021). An Empirical Study of the Influencing Factors on User Experience for Barrage Video Website—A Case Study of Bilibili. In V. G. Duffy (Ed.), Digital Human Modeling and Applications in Health, Safety, Ergonomics and Risk Management. AI, Product and Service. HCII 2021 (pp. 310–321). Springer.
Lommel, A., & Melby, K. (2018, March 17). A. MQM-DQF: a good marriage (translation quality for the 21st century). The 13th Conference of The Association for Machine Translation in the Americas.
Ramírez-Sánchez, G. (2022). Custom machine translation. Machine translation for everyone: Empowering users in the age of artificial intelligence (pp. 165–186). Language Science Press.
Sánchez-Gijón, P., & Kenny, D. (2022). Selecting and preparing texts for machine translation: Pre-editing and writing for a global audience. Machine translation for everyone: Empowering users in the age of artificial intelligence (pp. 81–104). Language Science Press.
Suojanen, T., Koskinen, K., & Tuominen, T. (2015). User-Centered Translation. Routledge.
Yang, Y. (2020). The danmaku interface on Bilibili and the recontextualised translation practice: a semiotic technology perspective. Social Semiotics, 30(2), 254–273.
Yang, Y. (2021). Danmaku subtitling: An exploratory study of a new grassroots translation practice on Chinese video-sharing websites. Translation Studies, 14(1), 1–17.

No competing interests reported.

Appendix.docx

A comparative study of machine translated subtitles based on the user-centered approach: a case study between Bilibili and YouTube

Status:

Version 1

Abstract

Figures

1. Introduction

2. Background

3. Methods

4. Data Analysis

5. Conclusion

Statements And Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1