Topic: “Exploring EFL Students’ Perceptions of the Semantic Depth in AI-Generated

Translations”
3.1. Definitions of Key Concepts
3.1.0. EFL Students’ perceptions
1. According to Woolfolk (1995), perception can be defined as “an interpretation of sensory
information” (p. 245). She elaborates that the word perception is defined as an opinion or belief
which is held by people on how they view something; this means that people can have different
perceptions based on their individual views.
2. Perception is the process by which the sounds of language are heard, assimilated, and
understood (Allotte, 2001).
3. Perception in the context of English learning includes how students understand,
assess, and respond to various elements of the language learning process, including
teaching methods, materials, and tools such as Google Translate. This perception plays an
important role in determining student motivation, engagement, and learning outcomes. As
Brown (2007) explains, students' perceptions of their learning experiences can influence how
they interact with the target language and how effectively they assimilate new language
skills.

Perception is a concept widely studied in psychology, linguistics, and education, with definitions
varying based on disciplinary context. According to Woolfolk (1995), perception is defined as
“an interpretation of sensory information” (p. 245), emphasizing its subjective nature. She
explains that perception reflects individual beliefs or opinions, shaped by personal experience and
viewpoints. This understanding implies that learners may interpret the same learning material
differently based on their cognitive and emotional filters.
Allotte (2001) approaches the concept from a linguistic perspective, defining perception as “the
process by which the sounds of language are heard, assimilated, and understood.” In this view,
perception is seen as an essential element in language processing, specifically focusing on
auditory comprehension. Compared to Woolfolk’s broader definition, Allotte’s is more narrowly
centered on sensory-linguistic processing—suggesting that perception, in the language context,
is not only mental but also input-based and skill-specific.
Brown (2007) offers a more educationally situated perspective, asserting that learners’
perceptions encompass how they “understand, assess, and respond” to different aspects of the
language learning process—including teaching approaches, learning materials, and digital tools
such as Google Translate. Brown emphasizes the impact of perception on motivation,
engagement, and learning outcomes, asserting that students’ subjective experiences influence
how effectively they acquire new language skills. This expands the notion of perception to
include cognitive, behavioral, and affective dimensions in EFL learning contexts.
When compared, all three definitions agree that perception is individualized and subjective,
playing a vital role in how learners engage with language input. However, they diverge in scope:
Woolfolk focuses on general cognition, Allotte limits perception to linguistic input, while Brown
frames it within pedagogical and technological contexts. Each perspective contributes to a
layered understanding of how students construct meaning from language learning experiences.
In this study, EFL students’ perceptions are operationally defined as:
“the learners’ individual and subjective interpretations of their English language learning
experiences, including how they perceive and evaluate tools such as AI-generated translations,
and how these perceptions influence their motivation, engagement, and language acquisition
outcomes.”

Woolfolk (1995) defines perception as “an interpretation of sensory information” (p. 245),
emphasizing its subjective and individual nature. Each learner may hold different opinions based
on how they view and experience the world.
Allotte (2001) narrows this concept within language learning, viewing perception as the process
through which language sounds are heard, processed, and understood—highlighting its sensory
and linguistic dimension.
Brown (2007), however, provides a more classroom-oriented perspective. He describes students’
perceptions as how they interpret and react to various aspects of the learning process, such as
teaching methods, materials, and tools like Google Translate. These perceptions directly
influence learners’ motivation, engagement, and outcomes.
Conclusion:
Based on these views, this study adopts Brown’s perspective and defines EFL students’
perceptions as:
“learners’ subjective interpretations of their English learning experiences, especially in relation
to AI-generated translation tools, which shape their engagement, motivation, and language
development.”

Comparison and Contrast:

All three definitions agree that perception is a mental process that varies among individuals.
However, they differ in scope and application. Woolfolk provides a broad psychological
perspective, applicable across various contexts. Allotte focuses on sensory-linguistic input,
limiting perception to language sounds and comprehension. Brown, in contrast, adopts a
classroom-centered approach, integrating perception with learning outcomes and the use of
digital tools. His view is particularly relevant to studies involving learners’ reactions to AI-
generated translation tools in EFL settings.

3.1.1. Semantic Depth

- **Mục tiêu**: Tìm hiểu và tổng hợp các định nghĩa về "semantic depth" từ ít nhất ba nguồn
khác nhau.
- **Hành động**:
 - **Tìm kiếm tài liệu**: Tìm các bài viết, sách hoặc nghiên cứu liên quan đến khái niệm
"semantic depth" trong ngữ nghĩa học.
- **Tổng hợp định nghĩa**:
- **Nguồn 1**: Trích dẫn định nghĩa và giải thích ngắn gọn về cách mà tác giả hiểu về
"semantic depth".
- **Nguồn 2**: Trích dẫn định nghĩa từ một tác giả khác, so sánh với định nghĩa đầu tiên (có
thể nhấn mạnh sự khác biệt hoặc tương đồng).
- **Nguồn 3**: Trích dẫn định nghĩa từ một nguồn thứ ba, tiếp tục so sánh và phân tích.
- **Đưa ra định nghĩa làm việc**: Dựa trên các định nghĩa đã tổng hợp, phát triển một định
nghĩa làm việc cho "semantic depth" phù hợp với ngữ cảnh nghiên cứu của bạn.
1. On measures of semantics, English exposure correlates with same-language performance on
receptive-expressive vocabulary or “lexical breadth” (Hoff et al., 2012, 2014), as well as indices
of semantic depth (Bohman et al., 2010). Here, “lexical breadth” refers to the stored lexical
entries that a speaker has; this roughly corresponds to words and includes nouns like árbol (tree),
phrasal verbs like “turn on,” and quantifiers like cada (every). This construct is typically assessed
using single-word vocabulary tests. “Semantic depth” comprises interconnected information
within a network that pertains to word meanings (i.e., definitions) and associations such as
synonyms and attributes.
2.
"Semantic translation attempts to render, as closely as the semantic and syntactic structures
of the target language allow, the exact contextual meaning of the original."
(Newmark, 1988, p. 39) Newmark, P. (1988). A Textbook of Translation. Prentice Hall.

Cruse, D. A. (1986) – Lexical Semantics Cruse, D. A. (1986). Lexical Semantics. Cambridge

University Press.
“The semantic depth of a lexical item involves the range and subtlety of its potential meanings,
which depend not only on denotation but also on connotation, collocation, and pragmatic usage.”
(Cruse, 1986, p. 57)

Baker, M. (1992) – In Other Words: A Coursebook on Translation

Baker, M. (1992). In Other Words: A Coursebook on Translation. Routledge.
“Translating is not just about transferring words from one language to another, but capturing the
semantic depth – the totality of intended meaning, cultural references, and rhetorical effect.”
(Baker, 1992, p. 21)

Semantic depth has been defined in various disciplines. Bohman et al. (2010) describe it as the
network of interconnected meanings around a word, including synonyms and attributes—
highlighting its role in vocabulary development. Cruse (1986) expands this view, emphasizing the
subtle layers of meaning shaped by denotation, connotation, collocation, and pragmatic use.
From a translation perspective, Baker (1992) sees semantic depth as the full richness of meaning
in a text, including cultural references and rhetorical effects. Similarly, Newmark (1988)
defines semantic translation as the effort to preserve the exact contextual meaning of the source
text.
Comparison and Contrast:
All four sources agree that semantic depth extends beyond surface-level meaning. Bohman et al.
(2010) and Cruse (1986) focus on intra-lexical relationships and contextual variation, while
Baker (1992) and Newmark (1988) apply the term more broadly to translation, stressing the
challenge of capturing deep and nuanced meanings across languages. Bohman and Cruse
emphasize word-level semantics, whereas Baker and Newmark shift toward text-level and
cultural-semantic fidelity in translation.
While Bohman focuses on word-level depth in language acquisition, Cruse and Baker address
deeper, context-sensitive meanings relevant to translation. For this study, Baker and Cruse’s
perspectives are most useful, as they align with the aim of exploring how EFL learners perceive
nuance, tone, and context in AI-generated translations.
Working Definition:
Based on these perspectives, this study defines semantic depth as:
“the extent to which learners understand the nuanced, context-dependent, and interconnected
meanings of words and expressions, including connotations, collocations, cultural references, and
rhetorical functions, especially as they appear in AI-generated translations.”

3.1.2. AI-Generated Translations

1. There is an increasing amount of research into ways to use machine translation to help students
read and write more successfully (Sefton-Green et al., 2016; Carrier, 2018; Reza, 2020). Using a
translator while studying a foreign language can help pupils write more fluently. Improve
communication skills and reduce errors. (Fredholm, 2019; Lee, 2020).
2. Machine translation (MT), traced back to the 1940s, refers to the automated process through
which computer programs and applications translate spoken or written texts from one language to
another. It has undergone several developments, from rule-based machine translation (RBMT) to
statistical machine translation (SMT) to the more revolutionized neural machine translation
(NMT) that is based on Natural Language Processing (NPL) (Rivera-Trigueros, Citation2022). It
pertained first in Google Translate’s 2016 release and has substantially improved the translation
quality (Omar & Gomma, 2020). Continuous advances in AI tools paved the way for Large
Language Models (LLMs) as a special type of deep learning that includes MT among the many
tasks it can generate and perform. ChatGPT is a groundbreaking language model that has gained
popularity worldwide since its release by OpenAI in 2022.

Machine translation (MT) has gained growing attention in language education for its ability to
support writing fluency, communication, and error reduction (Sefton-Green et al., 2016; Carrier,
2018; Reza, 2020; Fredholm, 2019; Lee, 2020). These tools, including Google Translate, have
shown positive effects on EFL learners’ performance and confidence.
Traditionally, MT refers to the automated translation of text via computer algorithms. As Rivera-
Trigueros (2022) explains, MT has progressed from rule-based (RBMT) and statistical models
(SMT) to neural machine translation (NMT), which uses natural language processing (NLP) for
better contextual accuracy. Google’s adoption of NMT in 2016 marked a major improvement
(Omar & Gomma, 2020).
More recently, large language models (LLMs) such as ChatGPT have extended MT capabilities.
Released by OpenAI in 2022, ChatGPT delivers context-sensitive and semantically rich
translations, representing a shift from basic translation toward more intelligent, user-aware
language generation.

Earlier studies highlight the educational benefits of MT tools in improving writing, fluency, and
learner confidence. In contrast, recent research focuses on the technological evolution from basic
rule-based systems to neural MT and AI-driven tools like ChatGPT. While traditional MT
prioritized accuracy and error reduction, AI-generated translations offer greater contextual
awareness and semantic depth—key to exploring how EFL students perceive meaning in
translated texts.
working definition:
In this study, AI-generated translations refer to language outputs produced by advanced artificial
intelligence systems—such as neural machine translation and large language models (e.g.,
Google Translate, ChatGPT)—that aim to deliver contextually accurate and semantically nuanced
translations for EFL learners.

- **Mục tiêu**: Tìm hiểu và tổng hợp các định nghĩa về "AI-generated translations".
- **Hành động**:
- **Tìm kiếm tài liệu**: Tìm các nghiên cứu, bài viết hoặc tài liệu mô tả về dịch thuật AI.
- **Tổng hợp định nghĩa**:
- **Nguồn 1**: Trích dẫn định nghĩa và giải thích từ một nguồn đầu tiên.
- **Nguồn 2**: Trích dẫn định nghĩa từ một nguồn thứ hai, so sánh với định nghĩa đầu tiên.
- **Nguồn 3**: Trích dẫn định nghĩa từ một nguồn thứ ba, tiếp tục phân tích.
- **Đưa ra định nghĩa làm việc**: Phát triển một định nghĩa làm việc cho "AI-generated
translations" dựa trên các định nghĩa đã tổng hợp.
3.2. Review of Related Studies
3.2.1. Tổng quan về các nghiên cứu liên quan
- **Mục tiêu**: Xem xét ít nhất 5 nghiên cứu cập nhật liên quan đến chủ đề của bạn.
- **Hành động**:
- **Chọn lựa nghiên cứu**: Tìm kiếm và chọn ít nhất 5 nghiên cứu gần đây liên quan đến nhận
thức của sinh viên EFL về độ sâu ngữ nghĩa trong dịch thuật AI.
 - **Tóm tắt từng nghiên cứu**:
- **Nghiên cứu 1**: Tóm tắt mục tiêu, phương pháp, và kết quả chính. Nhấn mạnh cách
nghiên cứu này liên quan đến chủ đề của bạn.
Siumarlata et al-2024
The purpose of this study is to analyze the semantic errors produced by online translation
programs (e.g., Google Translate, Bing Translator, Yandex, I-Translate), identify the challenges
they face, and propose ways to improve their accuracy and efficiency for better cross-cultural
communication.
Method: The research uses qualitative methods with a descriptive analysis approach, focusing on
sixth-semester English students at the Indonesian Christian University of Toraja. The study
consists of 10 Indonesian sentences translated into English using an online translation
application.
Findings: The study concludes that sixth-semester English students often make translation errors
when using online translation tools, particularly with context-specific terms, idioms, and
figurative language. These tools frequently fail to preserve the original meaning, nuance, and
style, leading to inaccurate translations.
Limitation: The study’s generalizability is limited due to the use of convenience sampling and
participants' similar backgrounds. Potential bias may have arisen from the researcher’s dual role
as instructor. Additionally, the 14-week duration may not capture long-term effects, and reliance
on self-reported data introduces the risk of social desirability bias.

Yijen – 2024
Aim: the present study aims to identify and classify the types of semantic errors made by MT in
translating students’ initial drafts written in Korean into English. It also attempts to make
pedagogical implications which can be implemented in L2 writing classrooms.
Method: The study used a mixed-methods approach with convenience sampling to investigate
how 79 Japanese EFL students in a compulsory course self-regulated their use of AI tools (e.g.,
ChatGPT, Google Translate) for writing, focusing on cognitive and sociocultural factors.
Ask ChatGPT
Findings: AI tools, specifically ChatGPT and Google Translate, enhance both the cognitive and
sociocultural aspects of the writing process. This confirms the cognitive theory perspective that
writing is a complex, mentally demanding process and aligns with sociocultural theory,
particularly Vygotsky’s concept of the Zone of Proximal Development (ZPD).
L: The use of convenience sampling and participants' similar backgrounds limit the
generalizability of the findings. The researcher’s dual role as instructor may have introduced bias,
and the 14-week study duration may not reflect long-term effects. The qualitative data rely on
self-reports, which may be affected by social desirability bias.
Zhang et al – 2025
This qualitative study, conducted in a researchintensive, Sino-foreign cooperative university in
southern China, explored university students’ perceived benefits and challenges of utilizing
GenAI in translation practices, as well as their preferred support mechanisms for addressing
encountered issues.
Method: Qualitative research open-ended questionnaires and semi-structured interviews
A research-intensive, Sino-foreign cooperative university located in Guangdong, China.
The study invited 136 university students enrolled in programs for translation and interpreting.
63 survey participants ( 18 males, 45 females) and 18 survey participants expressed their interest
in participating in the semi-structured interviews.
All participants achieved a minimum grade of B in courses related to translation.
10 participants were intentionally chosen to represent a range of backgrounds and characteristics.
F: Participants generally held positive attitudes toward GenAI tools due to benefits such as
enhanced translation efficiency, improved quality, feedback provision, diverse expressions,
motivation, and affordability.
However, they also expressed concerns regarding output inadequacy, prompting difficulties, lack
of reliability, transparency issues, and the risk of over-reliance affecting creativity and future
employment.
To address these challenges, students reported using self-initiated strategies and expressed a need
for teacher guidance and institutional support.
F: Participants generally held positive attitudes toward GenAI tools due to benefits such as
enhanced translation efficiency, improved quality, feedback provision, diverse expressions,
motivation, and affordability.
However, they also expressed concerns regarding output inadequacy, prompting difficulties,
lack of reliability, transparency issues, and the risk of over-reliance affecting creativity and
future employment.
To address these challenges, students reported using self-initiated strategies and expressed a
need for teacher guidance and institutional support.
L: Small sample from one Chinese university. Lack of longitudinal data.
Student perceptions may be influenced by instructors’ teaching approaches with GenAI.
Jaganthan et al - 2014
Aim: This paper focuses on the perception of GT usage among EFL learners in decoding
semanticity using GT in their learning environment. It aims at identifying the strategies taken by
the EFL student in selecting entries from Google Translator to arrive at a specific meaning.
M: Methods: qualitative, case study, questionnaire, interview
Samples: 9 international respondents from the University Sain Malaysia
Context: EFL student often use Google Translation in classroom
F: EFL learners heavily rely on Google Translate to understand word meanings, often selecting
the first entry provided without considering grammatical structure or contextual clues.
L: Short duration (7 weeks).
Limited sample diversity: the sample size was small (9 participants), and mostly from the Middle
East.
The study did not assess the accuracy of Google Translate, focusing instead on learners’
perceptions and strategies for selecting word meanings.
Tong King Lee et al-2023
This article asks how translation as a field of practice can best respond to this development.
Method: Qualitative analysis of translations produced by ChatGPT and other machine translation
(MT) systems.
Samples: Japanese-language passages from newspapers, literary texts, and celebrity biographies.
Context: A comparison of translation performance between ChatGPT and well-known MT tools,
within the broader context of AI's growing influence on the translation industry.
F: ChatGPT performs competitively or even better than other MT systems, especially in
discourse-level translation (e.g., pronoun consistency, coherence).
Prompting significantly improves ChatGPT’s performance.
ChatGPT cannot fully replace human translators but can function as a powerful tool.
The translator’s role is shifting from linguistic executor to language consultant or project
manager.
Limitation: ChatGPT results are inconsistent across prompts and iterations (non-replicability).
AI lacks human understanding, contextual inference, and ethical responsibility.
Evaluation is time-sensitive—performance may vary with newer versions of the model.
1. Overview of Reviewed Studies
 5 recent studies from 2014–2025 related to AI-generated translations and EFL learners.
 Focus areas: semantic errors, learner perceptions, GenAI tools (e.g., ChatGPT, Google
Translate).

🧠 2. Main Findings
 Siumarlata et al. (2024):
MT tools struggle with idioms, figurative language, and context, leading to semantic
inaccuracy.
 Yijen (2024):
Students use AI tools for writing support but still rely on teachers and peers. AI supports
cognitive engagement, but not fully independent learning.
 Zhang et al. (2025):
Students appreciate GenAI tools for efficiency and quality, but worry about over-
reliance, prompting, and creativity loss.
 Jaganthan et al. (2014):
Learners often accept first translation result without checking contextual accuracy →
shallow semantic engagement.
 Tong King Lee et al. (2023):
ChatGPT performs well in discourse-level coherence but lacks human-like inference
and ethical responsibility.

🧪 3. Methodologies
 Mostly qualitative (interviews, open-ended surveys).
 One mixed-methods study (Yijen, 2024).
 Many rely on self-reporting → potential bias.
 Short research durations limit long-term insights.

🌏 4. Research Contexts
  Conducted in Indonesia, Japan, China, Malaysia.
 Focus on university EFL students.
 Mostly Asian, non-Western contexts.
 Studies centered on translation classes or writing tasks using AI.

❗ 5. Research Gap
 Few studies explore how learners perceive semantic depth in AI translations.
 Most focus on translation errors or tool usage, not meaning interpretation.
 Need to understand how learners judge nuance, tone, and context in AI-generated
output.

- **Nghiên cứu 2**: Tóm tắt tương tự như trên, so sánh với nghiên cứu 1 (có thể là về phương
pháp hoặc kết quả).
- **Nghiên cứu 3**: Tiếp tục tóm tắt và so sánh.
- **Nghiên cứu 4**: Tóm tắt và phân tích.
- **Nghiên cứu 5**: Tóm tắt và so sánh.
- **Tổng hợp và phân tích**:
- **Synthesize findings**: So sánh các kết quả của các nghiên cứu, tìm ra các xu hướng chung
hoặc khác biệt.
- **Synthesize methods**: Phân tích các phương pháp nghiên cứu được sử dụng, nhấn mạnh
những điểm mạnh và điểm yếu.
- **Synthesize contexts**: Xem xét các bối cảnh khác nhau mà các nghiên cứu được thực
hiện, và cách điều này ảnh hưởng đến kết quả.
- **Nhấn mạnh nhu cầu cho nghiên cứu của bạn**: Dựa trên các phân tích trên, chỉ ra khoảng
trống trong nghiên cứu hiện tại và lý do tại sao
nghiên cứu của bạn là cần thiết.