SHS Web of Conferences 163, 02021 (2023) https://doi.org/10.

1051/shsconf/202316302021
ICSSED 2023

The Impacts and Challenges of Artificial Intelligence Translation

Tool on Translation Professionals
Lan Wang*
Shandong Jianzhu University, Jinan 250101, Shandong Province, P. R. China

Abstract: Machine translation, especially translation based on neural network technology has made a major
breakthrough and is increasingly accepted and widely used. The development of artificial intelligence (AI)
translation has had a definite impact on translation jobs. People, even professional translators, are relying on
AI translation. But There is no research on whether machine translation software is superior to professional
translators in translating various types of documents.. In this study, we design an experiment to determine the
advantages and disadvantages between AI translations and human translations. The result shows the impact
of the development of AI on the translation industry. To achieve better translation results and output high-
quality translations in the era of rapid development of AI, it will contribute to Human-AI partnerships.

and other methods to investigate this area.

1.INTRODUCTION
2.Literature Review
With the development of artificial intelligence technology,
the core of computer aided translation is machine This experiment will be conducted using the principles of
translation and translation memory [1]. With its the Turing Test [6], in which the translations produced by
continuous development, translation tool like Dear the AI tool are disordered with those made by human
Translate1, Baidu Translate2, DeepL Translate3 are widely translators without revealing information about the
used and influential in the translation field, recognized by translators, and the research subjects are asked to rank the
people. Although AI translation is widely used as an quality of the translations. This will determine whether the
essential tool, and it brings great convenience to people, quality of the translations produced by the AI translation
the phenomenon of people relying on AI translation has tool exceeds that of the human translations, yielding
become more and more serious, has caused translation results on the development of AI and its recognition in
positions to be impacted to a certain extent, and there are people's minds.
pros and cons to professional students’ training [2]. Many scholars believe that the existing translation
With industry characteristics as the basis for training standards and translation criticism standards are too
English language talents in different directions to meet the subjective and lack objectivity and an index system [7].
needs of society, the training of language professionals in However, judging translations in terms of accuracy to the
colleges and universities is gradually focusing on other original text, linguistic style, the connotation of the source
areas of training. We can call them specialties of high-end language, ideology, cultural background, and logical
complex talents on English [3]. Language translation is an expression is now recognized by many English
important part of English language education and is an professionals[8]. A word may have multiple meanings in
essential component of professionals' competencies. the translation process, so the translator needs to choose
Culturing this competence usually requires the the best meaning that is faithful to the text, acceptable to
combination of theory and practice, such as using the the reader, and conveys the connotation and emotion that
theoretical knowledge acquired to translate works to the author wants to convey. This means that translations
improve one's translation skills. However, some English need to meet the translation standard of Faithfulness,
professionals still rely on AI translation in their translation Expressiveness, and Elegance needs to be satisfied in the
practice, which is not conducive to developing their translation process.
English language skills[4]. In recent years, some Therefore, in this study, through a comparative
researchers have suggested that the rapid growth of AI experiment, English professionals and non-English
translation tool could hurt English professionals [5]. professionals were selected as the study subjects, and they
Therefore, this paper will use comparative experiments will evaluate different translations from the judgmental

1 https://fanyi.youdao.com/ 3 https://www.deepl.com/translator
2 https://fanyi.baidu.com/

*Corresponding author: [email protected]

© The Authors, published by EDP Sciences. This is an open access article distributed under the terms of the Creative Commons Attribution
License 4.0 (http://creativecommons.org/licenses/by/4.0/).
SHS Web of Conferences 163, 02021 (2023) https://doi.org/10.1051/shsconf/202316302021
ICSSED 2023

perspectives mentioned above. English-speaking texts, News reports and Literary works. Each of them
professionals are more expert in judging translations from contains three passages which contain six translations
different perspectives and criteria. Other non-specialists respectively. Half of the six translations are human
have also commented on the convenience of translation translations, and half are AI translations. (See Figure 1)
tools as a tool and the acceptability of the translation to
them.

3.Research Design

3.1.Types and characteristics of articles Figure 1 An example of one type of article

This study will conduct a comparative experiment to Step 2: The judges scored the six translations
select different articles, namely Business texts, News according to the rubric. The scores scored by all subjects
reports, and Literary works. These articles have various were averaged and ranked in descending order according
linguistic features. Business texts are mainly used in daily to the average score.
work, mainly for daily social interaction. They emphasize Step 3: Take the top three ranked translations and
the accuracy of language and the accuracy of technical compare them. The translation with the higher percentage
terms. News articles are widely drawn and need to reflect of the top three will win in this passage.
accuracy and timeliness. Literary works mainly reflect the Step 4: Continue the process from step 3 three times to
different translation styles of language and the need to find out who has won in each passage. The one with the
understand the connotation of the text for translation. It is highest number of wins among the three passages wins in
a good example of manual translations and AI translations. this type of article.
Step 5: The one with the highest number of wins
among the three types wins the final. (See Figure 2)
3.2.Evaluation criteria
After selecting three articles according to the above
principles, select three passages in each type of article and
choose six translations for them. The six translations are
classified as standard translation, high-level artificial
translation, low-level artificial translation (for freshmen
English majors), Dear Translate’s translation, DeepL
Translate’s translation, and Baidu Translate’s translation.
Then these translations are out of order without Figure 2 Experimental rules
revealing the translator's information, so that English
professionals or non-English professionals can choose. 4.Findings
Let them rank the quality of translation according to
translation criteria. The translations are evaluated in six
areas: accuracy of the original text, linguistic style, 4.1.Experimental results of different types
connotations of the source language, ideology, cultural
context, and logical expression. During the experiment, 4.1.1Experimental results of business texts
we can ask the experimenter whether they need to change
the answer by misleading language so that they can make The respondents in this experiment were scored according
the answer after careful consideration and avoid the to three criteria and each translation was ranked from
chance of the experiment. After the selection is completed, highest to lowest. In the third type of article, the human
conduct relevant analysis, and inform them of the results translator finally won, and the AI won only once out of
for personal interviews to talk about the views of AI nine passages. For example, in Type 1, two of the top three
translation. translations in passages 1.1 under this type were AI
translations, so the AI won in round 1. However, in
passages 1.2 and 1.3, two of the top three translations were
3.3.Experimental rules
human translations, which means that the human won in
This experiment is divided into five steps: these two passages. Therefore, the overall score was 2:1,
Step 1: there are three types of articles, namely Business with the human winning in this type of text, i.e., business
English translation. (See Table 1).
Table 1 Experimental results of type1
Average Round 1 (In
Points (Full passages)
#Passag 15 points) Round 2 (In
Order (Top three out of six translations)
e type 1)

TOP1 Dear Translate 12.2 AI wins (2:1)

Human wins
1.1
TOP2 Baidu Translate 11.8 (2:1)

2
SHS Web of Conferences 163, 02021 (2023) https://doi.org/10.1051/shsconf/202316302021
ICSSED 2023

High-level artificial 10.6

TOP3 translation
TOP1 DeepL Translate 11.2 Human wins
(2:1)
Low-level artificial 9.8
1.2 TOP2 translation
TOP3 Standard translation 9.2

TOP1 Standard translation 12.6 Human wins

(2:1)
Low-level artificial 11.8
1.3 TOP2 translation

TOP3 Dear Translate 11.8

For the note is that in the first passage 1.1 of Business and passage 1.3 (0.38). This also shows that DeepL
English, Dear Translate and Baidu Translate take the top Translate is much better than human translation in this
two places and score much higher than human translators. translation.
We can also see from these three translations that Dear
Translate appears more frequently in the translation of
4.1.2Experimental results of news
business English. This shows that Dear Translate is
machine translators' most popular translation tool. In In the second type of article, in the end, the humans won
addition to this, in passage 1.2, the top three translations with a score of 3:0. (see Table 2)
scored a discrete 0.84 (higher than it in passage 1.1 (0.68)
Table 2 Experimental results of type2
Average Points Round 1
Order (Top three out of six Round 2 (In type
#Passage (Full 15 points) (In
translations) 2)
passages)
Standard 11.6 Human
TOP1
translation wins
Low-level 11.4 (2:1)
2.1 TOP2 artificial
translation
Dear 11.2
TOP3 Translate
Standard 13.8 Human
TOP1 translation wins (3:0)
High-level 12.4
2.2 TOP2 artificial
translation Human wins
Low-level 11.4 (3:0)
TOP3 artificial
translation
Standard 12.6 Human
TOP1
translation wins (3:0)
High-level 11.4
TOP2 artificial
2.3 translation
Low-level 11.0
TOP3 artificial
translation

From passage 2.1, although the AI translation came in

third, it scored highly at 11.2. And the dispersion of the top
4.1.3Experimental results of literature works
three was 0.16, the smallest of all the discrete data. This
suggests that in this translation, Dear Translate's Similarly, in the literature theme, we can see that in
translation is not as professional as the standard translation, passages 3.2 and 3.3, the top three are all human
but the experimental subjects also recognize the translations. Among the texts in this category, only DeepL
translation. In both passages 2.2 and 2.3, the top three were Translate's translation was ranked in the top three with a
all human translations. Thus, we can see that in score of 11.6 in passage 3.1. Thus, in the literature
translations of news topics, AI is not as good as human category, human translations are more likely to be
translation due to the need for rigorous logic, and accuracy accepted by readers, as humans can better appreciate the
and conciseness of language wording in this category. meaning of the text and the author's emotions. (See Table
3)

3
SHS Web of Conferences 163, 02021 (2023) https://doi.org/10.1051/shsconf/202316302021
ICSSED 2023

Table 3 Experimental results of type 3

Average Round 1 (In
Points (Full passages)
Order (Top three out of six
#Passage 15 points) Round 2 (In type 3)
translations)

Low-level 12.4 Human wins

TOP1 artificial (2:1)
translation
3.1 High-level 11.8
TOP2 artificial
translation
TOP3 DeepL Translate 11.6
High-level 12.0 Human
TOP1 artificial wins (3:0)
translation
Standard 10.6
3.2 TOP2 Human wins
translation
(3:0)
Low-level 10.0
TOP3 artificial
translation
High-level 12.0 Human
TOP1 artificial wins (3:0)
translation
Standard 11.2
3.3 TOP2
translation
Low-level 10.0
TOP3 artificial
translation

works and emotive texts. Secondly, in terms of Logical

Expression and Fidelity to the original text, and accurate
4.2.The results of AI translation in three
translation of word meanings, we can see that humans win
criterions
all three types of texts, i.e., 3:0. In other words, in these
According to the experimental rules we designed, the areas, AI is still lacking, and needs to be improved.
translations were scored according to the criteria for
judging the translations (see Table 4). 4.3.Respondents' feedback
Table 4 Analysis of evaluation standards
Fidelity to According to our interviews with the respondents, they
the original believe that AI translation has problems with the choice of
text and context and the logic of the language and still needs to be
Logical Language
Types accurate improved. They believe that AI translation is now
expressions style
translation mechanized and that it is no longer able to analyze the
of word deeper meaning of the author's message, leading to errors
meanings in translations. However, the respondents still have a
Type 1 Human wins Human wins AI wins positive attitude toward the development of AI translation.
They also often use translation tools when consulting
Type 2 Human wins Human wins AI wins English literature, making their learning more efficient. So
Human they hope that translation professionals will be able to
Type 3 Human wins Human wins collaborate manually with AI to ensure efficiency and
wins
improve the accuracy of their translations, complementing
Final Human Human wins AI wins each other's strengths and weaknesses.
winner wins (3:0) (3:0) (2:1) They were not surprised by the human victory of the
As we can see from table 4, AI is better at grasping experiment. But they still believe that the rapid
language style than humans. This is mainly the case for development of AI has already seen some translations
business English translations and news types. This also approach the level of human translation, so they support
shows that for texts requiring a strict language style, AI the development of AI translation tools. Some of them also
translations avoid colloquial expressions and are better realize the rapid growth of AI, which they believe is
than human translations. However, AI translations are not already a danger to their profession. They are therefore
as good as human translations for the translation of literary more concerned about issues such as unemployment and
how to improve their own before, they are more concerned

4
SHS Web of Conferences 163, 02021 (2023) https://doi.org/10.1051/shsconf/202316302021
ICSSED 2023

about issues such as unemployment and how to improve them as translators to have a profound grasp of the
their core competencies in an era of rapid AI development. relationship between the source language and the target
language, and for the translation to be faithful to the source
language while allowing the target language readers to
5.Conclusions understand the original text. Today's AI translation tool is
developing rapidly, but it still needs to be upgraded to
5.1.Refinements needed AI translation to meet the standard of "faithfulness and elegance[9]."
improve in some areas Therefore, at this stage, human-machine collaboration can
be used under the principle of translation efficiency.
Because of the abstract and metaphorical nature of the We should first see the benefits of AI. People gradually
Chinese language, AI translation needs to be improved in adapt to the convenience brought by AI translation, and
some areas. The first is about logical expression. English even AI promotes the development of English major
is linear, whereas Chinese is spiral and diverse in form. learning, which has been widely recognized by many
That is to say, English emphasizes the subject-predicate teachers and students in practical application [10].
object, whereas Chinese sometimes does not need a However, for the future translation industry, AI translation
subject or predicate to form a sentence. Apart from this, tools should reflect its instrumental nature[11]. Using AI
the logical order of language expressions is also different. to efficiently translate the original text into the first
For example, English prefers to use the passive voice, version of the translation. In view of the content biased to
emphasizing the object, while Chinese prefers to use the culture or emotion in the article, we humans, as animals
active voice, emphasizing the subject. In translation, some with feelings, should learn more about the connotation of
logical words in English are omitted because the Chinese the text, proofread the AI translation, correct translation
language usually indicates the logical relationship by the errors, ensure smooth logic and accurate word meaning
order of the sentences. This requires the study and analysis selection, and finally output high-quality translation to
of sentences. In this regard, human beings have been achieve artificial cooperation.
trained to think a great deal. It is easier to accurately grasp
the logical relationships in sentences when translating
from the source language into the target language. 6. Conflicts of Interest
Secondly, there is also a need to improve the fidelity of
The author declare that the study is done and responsible
the original language. In this experiment, we can see that
for the experimental results.
of the three human-translated translations in each passage,
they were all more accurate than the AI translations. We
can conclude that humans can almost avoid mistakes with References
the help of relevant tools such as dictionaries. However,
suppose a human seeks to translate a text with a relatively 1. Luo Qichen, Jin Enguang & Li Yafen (2022). A survey
high accuracy rate in a short period without the use of any on computer-aided translation software under the
tool, e.g., a dictionary. In that case, he or she may be background of AI--based on Youdao Dictionary. The
unable to beat an AI translation. We can therefore see that Computer Age, (8), pp.43-45, 55.
there are still advantages to artificially intelligent 2. Okpor, M. D. (2014). Machine translation approaches:
translation. It is more efficient than human translation and issues and challenges. International Journal of
has higher accuracy in grasping the content of the text. Computer Science Issues (IJCSI), 11(5), 159.
In addition to this, we can see that in terms of language
3. Mi Fenfang & Xu Ningjun. (2022). Employment
style, AI translations are already readable and acceptable
Challenges and Countermeasures for English
to readers. In business English, for example, the language
Majorsin Universities under the New Situation.
used in AI translations is also very concise and formal and
Journal of Qiqihar University (Philosophy and Social
avoids colloquialisms to a large extent. Here we can see
Science Edition), (10), pp.149-151.
that AI is becoming increasingly sophisticated in language
translation and is facing a significant impact on English 4.Bogong Jia. (2022). The Impact of Using “Machine
translators. Translation” on Chinese CollegeStudents’ English
Writing. [Master, Jilin University]. Jilin University.
5. Gu Yue & Zhang Jiayuan. (2020). The pros and cons of
5.2.Possibilities for collaborative human
machine translation software for university students
translation in a fast-paced period
learning English. English Square, (28), pp.115-117.
English professionals and AI translators are 6. French, R. M. (2000). The Turing Test: the first 50
complementary. As a bridge between the original author years. Trends in cognitive sciences, 4(3), 115-122.
and the reader, the translator can dissect the connotations 7. Rosemary, M.S. (2015).A reconsideration of translation
of the text and thus sort out its logical structure, thus quality and standards. Журнал Сибирского
compensating for the problems of AI translation at this федерального университета. Гуманитарные науки,
stage. Whereas there are problems associated with the 8(12), 2908-2919.
translator, AI translators can analyze through algorithms
8. Mandal, P. C. (2018). Translation in qualitative studies:
the basic grammatical errors as well as the cost of time.
Evaluation criteria and equivalence.The Qualitative
For English professionals, who themselves have mastered
Report,23(10), 2529-2537.
two or more languages, it is all the more important for

5
SHS Web of Conferences 163, 02021 (2023) https://doi.org/10.1051/shsconf/202316302021
ICSSED 2023

9. Zhang, M. (2013). An Inquiry into Yan Fu's Translation

Theory of Faithfulness, Expressiveness, and Elegance:
The Beginning of China's Modern Translation
Theory. Trans-Humanities Journal,6(3), 179-196.
10. Cao, L., & Zhu, S. (2022). Analysis of the Factors
Influencing the Adaptability of College English
Learning Based on AI Teaching Assistance.
Mathematical Problems in Engineering, 2022.
11. Folaron, D. (2010). Translation tools.Handbook of
translation studies, 429-436.