Analysis of high-frequency errors and linguistic patterns in EFL medical students’ English writing: Insights from a learner corpus

Abstract

Background

The perceived language barrier in English is said to hinder, and in certain instances, impede the global dissemination of knowledge, including medical information, to non-native English speakers within medical institutions. As English for medical purposes instructors, we contend that the issue persists in medical universities across various EFL contexts. Medical students face the challenge of presenting their research findings in English for international journals and conferences. Given this, the present research study aimed to compile a comprehensive catalog of high-frequency errors and examine them in recurring linguistic patterns commonly found in the writing of Iranian medical students.

Methods

In conducting the present study, we developed a learner corpus of 1,040 essays (339,040 words and 18,235 sentences in total). Through using the results obtained from Wordsmith Tools 8 and sifting the leaner corpus, we identified 11 high-frequency errors and five commonly used linguistic patterns.

Results

Only five out of 11 high-frequency errors account for 61% of the total number of errors. Results also showed that a majority of errors were of grammatical nature. In this regard, cohesion and cohesive devices (16%) were the most prevalent errors followed by omission/misusing of articles/determiners (14%). Additionally, results showed that discourse markers were extensively used in the corpus (22.07%), followed by hedges (11.42%).

Conclusions

The outcomes of this study are expected to assist English for medical purposes instructors in designing focused lesson plans and classroom activities. Ultimately, these efforts might contribute to enhancing medical education in non-English speaking universities.

Background

The English language barrier is claimed to restrict or, in some cases, even block the globalization of knowledge including medical resources for English non-native speakers at medical schools and universities [1]. Emphasizing writing as an important skill that makes publication in English a demanding process, Osler and Starkey (2015) highlight challenges faced by English non-native learners including medical students [2]. In the same vein, we as English for medical purposes (EMP) instructors, argue that the problem continues to exist at medical universities in several EFL contexts, challenging medical students to report the findings of their research studies in English in international journals and conferences. Given this, we decided to conduct the present study to identify the high-frequency errors and patterns in EFL medical students’ English writing. Study findings were expected to contribute to breaking down language barriers and enhancing the global dissemination of medical knowledge, as well as improving medical education in non-English speaking universities.

Considering the test-based instruction in several EFL contexts, teaching writing has received less attention in English language classes, as a bulk of final exams is in the multiple-choice format. Our impetus for the present study originates from the idea that error analysis allows us to thoroughly investigate the errors made by EFL medical students, identify high-frequency patterns, and pinpoint areas of weakness that require further attention. Also, to the best of our knowledge, there is no study investigating EFL medical students’ high-frequency errors and linguistic patterns in English writing. To address the identified goal and gap, this study was conducted with the expectation that its findings will also be applicable to other similar contexts beyond Iran in teaching EMP to EFL medical students.

Error analysis

An error in language refers to the usage of linguistic elements in a way that is perceived by a native speaker as indicative of faulty or incomplete learning [3]. Alongside, Keshavarz (2008) argues that a mistake is a performance error that is either a random guess or a slip, while an error is a systematic deviation made due to a lack of knowledge of the correct rule [6]. Errors can be categorized into three types, as proposed by Richards and Schmidt: lexical errors (related to vocabulary), interpretive errors (resulting from a misunderstanding of the speaker’s intention), and pragmatic errors (creating an incorrect communicative effect). Another perspective on error sources distinguishes between interlingual errors (arising from the learner’s first language) and intralingual errors (stemming from incomplete or flawed learning of the target language) [4].

Additionally, the error analysis methodology [5, 6] which serves as the basic framework for the present research consists of three stages: (a) error recognition and reconstruction, (b) error description, and (c) error explanation. The first stage involves the identification of errors in the corpus. In the second stage, errors are systematically classified into various categories such as grammatical and lexico-semantic, detailing specific types such as cohesion and cohesive devices, omission/misuse of articles/determiners, etc. The third stage of the methodology centers on explaining the underlying systems that give rise to these errors. Drawing inspiration from this method, the present study utilizes a modified version, which will be further discussed in the Findings section of this research.

Learner corpus and error analysis

One growing field in corpus linguistics is the learner corpus, which allows researchers to compare learners’ authentic language with that of native speakers [7, 8, 11, 12]. Learner corpora are particularly useful in error analysis, as they provide insights into common learner errors and their frequencies. For example, Tono et al. (2014) used a learner corpus to identify frequent omission and addition errors among learners, finding that these errors were effectively addressed using targeted interventions [9]. Additionally, Bridle (2019) highlighted that learners often used corpus tools to correct wrong word and formal/informal errors, although they made limited use of these tools for other types of L2 writing errors [10].

Precursory studies on the error analysis and high-frequency linguistic patterns

Results of the line of research on error analysis suggest that grammatical errors are the most frequent type of error, followed by lexical errors [14–16]. In this regard, errors related to the sentence structure and verb forms are particularly prevalent in EFL students’ writing [13, 17]. Also, Sasi and Lai (2021) reported misformation errors (i.e., related to wrong morphemes and the sentence structure) as the high-frequency error type among Taiwanese university students [15]. Additionally, Zhan (2015) observed that errors in tenses were the most common among Chinese EFL students’ writing [18]. Also, L1 transfer was seen as a major contributing factor to errors in L2 writing [19, 20]. In the same vein, Bao (2015) argued that mother tongue interference, insufficient comprehension, and application of L2 rules were the main sources of errors in L2 writing among Chinese EFL learners [21].

In addition, effective writing necessitates skillful utilization of diverse linguistic patterns, including discourse markers (connectors that establish links and organize ideas within a text, such as also, but, since, etc.) and meta-discourse markers (devices that convey the writer’s perspective and attitudes towards the discourse, namely central modals, informal features, hedges, and boosters)whose incorrect use may produce a gap in the communication between the writer and the reader. Extensive research has focused on the frequency and correct usage of discourse markers in learners’ writing. Findings of precursory studies [22–27] indicate that the use of elaborative discourse markers, which signify expanded information, is prevalent. Also, errors often occur in relation to conjunctive devices, which are commonly associated with the utilization of discourse markers. Martinez (2004) attributes the heightened occurrence of elaborative markers (e.g., and, moreover, furthermore) to the demand for elaboration in expository compositions, often signaled by the presence of these markers [25].

Additionally, a considerable amount of research has been conducted on the use of two meta-discourse markers, hedges and boosters. Hedges are used to show a degree of uncertainty regarding the truth value, while boosters serve to demonstrate certainty [28]. These markers have been examined in diverse genres using various comparison criteria. Certain studies have investigated gender differences in the utilization of hedges and boosters [29], while others have focused on comparing meta-discourse marker usage across different academic disciplines [30, 31]. Also, some studies have explored differences in the use of hedges and boosters across different cultures [32].

Although Crismore et al. (1993) and Grabe and Kaplan (1997) argue that hedges and boosters are interrelated concepts [32, 33], several researchers have conducted studies focused solely on the hedging concept (e.g., 38–45). Conversely, research addressing only the concept of boosters has been limited (e.g., 46–48). Furthermore, numerous studies have examined the functions of discourse markers [35–38] and meta-discourse markers [28, 39–42].

There has also been a widespread assumption in recent years that the writing style employed in hard science disciplines (such as medicine) has undergone a shift towards informality [43–46]. In this regard, scholars have focused on certain lexical and grammatical characteristics that are typically associated with informality. Examples include the use of first-person pronouns [47–51], imperatives [52], pronominal anaphoric references [53–55], contractions [56], and sentence-initial conjunctions [57]. Furthermore, some studies have grouped these informal features together and analyzed their overall occurrence as indications of varying degrees of informality [46, 58].

While few studies have investigated central modals in EFL writing, no research, to the best of our knowledge, has yet explored this in the context of EMP. Central modality encompasses a range of semantic concepts, including necessity, probability, ability, intention, obligation, and hypothetical scenarios, collectively referred to as modal meanings [59]. In a corpus-driven study, Elturki and Salsbury (2016) examined the development of English central modality in the written narratives of Arab EFL learners across six proficiency levels. Their findings indicate that Arab learners begin to express modality concepts before acquiring target-like linguistic patterns. At lower proficiency levels, learners predominantly utilize salient forms characterized by high frequency and low variation. The authors attribute this pattern to the influence of the EFL learners’ native Arabic modal system [60].

Although precursory studies have elucidated the errors and linguistic patterns in the writing of both EFL learners and native English speakers, to the best of our knowledge, no research has specifically focused on the high-frequency errors in the English writing of EFL medical students, nor has it examined the prevalence of such errors within high-frequency linguistic patterns in their writing. Furthermore, a notable limitation of prior investigations lies in the relatively small corpus size utilized for analysis. In this regard, in social sciences, the valid sample size of a corpus has always been under controversy. It has also been argued that the size of a corpus is an essential principle [11, 61, 62], as smaller corpora may not sufficiently represent the subject under the study. The literature in this field indicates that only three error analysis studies [15, 63, 64] have been conducted using learner corpora exceeding 200 essays. Specifically, Chan (2010) analyzed 696 essays written by ESL learners in Hong Kong, Hart (2017) examined 600 essays by Chinese EFL learners, and Sasi and Lai (2021) investigated essays from 430 EFL Taiwanese university students. Other error analysis studies were conducted with fewer essays (Lan et al. (2019): 100 essays; Pimpisa and Normah (2015): 40 essays; Zhan (2015): 177 essays; Zheng and Park (2013): 168 essays) [16, 18, 19, 65].

Research questions

Ellis argues that error analysis helps instructors in identifying learners’ problems at any phase of their language development [66]. By analyzing EFL medical students’ writing through a learner corpus, we can better grasp the linguistic challenges they are faced with in English writing, which, in turn, helps EMP instructors adapt classroom strategies in order to address medical students’ high-frequency errors in commonly used linguistic patterns. To be more specific, we aimed to explore the following questions in the present study:

1. What are EFL medical students’ high-frequency errors in English writing?

2. What are linguistic patterns that EFL medical students commonly use in their English writing?

3. What is the percentage of high-frequency errors in each of these commonly used patterns?

Methods

The aim and design of the study

In this research, our primary objective is to compile a comprehensive catalog of high-frequency errors and examining these errors in recurring linguistic patterns found in the writing of Iranian medical students. To achieve this, we employed the text analysis method, specifically by constructing a learner corpus which consists of essays authored by EFL medical students.

The rationale for examining high-frequency linguistic patterns within the identified instances of errors lies in the interconnected nature of language use and error production. By analyzing both aspects within a single study, we can identify potential correlations between specific patterns and error types. This can provide insights into whether certain linguistic patterns are more prone to errors, which can inform targeted instructional strategies [66]. Also, language use is inherently multifaceted, and separating error analysis from linguistic pattern analysis might overlook the contextual factors that contribute to errors. A combined analysis offers a holistic view of students’ writing, capturing the interplay between correct and incorrect usage [6]. In addition, utilizing a single dataset for both analyses ensures consistency and maximizes the value derived from the collected data. Figure 1 represents a flowchart of the research Methods. Also, the present study was approved by the Ethics Committee of Iran University of Medical Sciences (IR.IUMS.REC.1403.120), and written informed consent was obtained from all participating students.

Fig. 1.

Flowchart of the research Methods and Results

The participants

Participants of the present study were the intermediate-level EFL medical students taking the General and Technical English language courses at Iran University of Medical Sciences (IUMS), as we believe that students at this proficiency level are able to generate a variety of grammatical structures and lexical items due to the opportunities provided in the courses. The participants were supposed to be at the intermediate English language proficiency level according to the results of their English language placement test taken at the beginning of the semester. Additionally, ongoing class formative assessments further confirmed their intermediate English language proficiency level. Of the 1,040 participants in this study, 573 were female and 467 were male at the age range of 19–25.

Developing and analyzing the learner corpus

To develop a learner corpus, we conducted a writing activity through which participants were required to write on a medical topic (see appendix 1 for the writing activity). The writings were collected in the hard copy. In doing so, we first provided the participants with the writing prompt, and explained it in detail to make sure that the participants well understood what was expected of them and what they needed to include in their writing. The participants were to write an essay in no fewer than 250 words in a single session of 90 min; this way, they had enough time for brainstorming and drafting at their own pace. One of the authors conducted the sessions and collected the essays from April 2024 through July 2024. Also, all the EFL medical students participated voluntarily in the study and signed the consent form at the beginning of the session.

We collected a total of 1,040 essays, each containing between 250 and 423 words. Subsequently, we created a learner corpus comprising 339,040 words and 18,235 sentences. To analyze this corpus, we initially converted the essays into Word document files using the Google Docs verbatim converter. Additionally, we addressed any typos that remained from the Google Docs conversion process. In the next step, we used Wordsmith Tools 8 [67] to analyze the learner corpus in terms of types and frequency of errors and linguistic patterns. Wordsmith Tools 8 is a widely used software by linguists and language researchers for corpus data analysis [68, 69]. It includes several modules such as Concord, WordList, and KeyWords. The Concord module generates concordances, allowing us to see how words are used in context. The WordList module creates frequency lists of words or word clusters, which helped us identify common errors and patterns. Also, the KeyWords module identifies words that are unusually frequent or infrequent in comparison to a reference corpus, providing insights into distinctive linguistic features of the learner corpus [67].

Error analysis method

We adopted a modified version of the error analysis method that includes developing the learner corpus, identification of errors, and classification of errors. The rationale behind the utilization of this modified version lies in the value of delving into the underlying systems of errors and assessing their significance (error explanation) [5, 66, 70].

After making the learner corpus, the next step involves error identification, which is accomplished through initial proofreading of collected and anonymized essays. The next stage includes error classification and quantification, where errors are categorized and counted. We then quantify the overall frequency of errors in the learner corpus. Given the numerous errors in the corpus, we focus our analysis on high-frequency errors, defined as those occurring at least 100 times.

Results

The Results section is divided into two parts. In Part 1, we address the first research question by developing a comprehensive list of high-frequency errors in English writing among EFL medical students. In Part 2, we respond to the second and third research questions by identifying high-frequency linguistic patterns and calculating the percentage of high-frequency errors within each pattern.

High-frequency errors in EFL medical students’ English writing

After reviewing the data obtained from Wordsmith Tools and manually sifting through the learner corpus, we detected 9,647 errors. The error identification process involved both automated analysis using Wordsmith Tools and a thorough manual review by three grammar experts. Each expert independently examined the corpus to identify errors, ensuring a comprehensive detection process. To ensure reliability, 89% of the detected errors were reviewed and confirmed as errors by all three experts. For the remaining 11% of errors where there was no consensus, we conducted a secondary review process to resolve discrepancies. Only errors confirmed by at least two experts were included in the final analysis, resulting in a total of 8,691 errors, averaging eight errors per essay. This approach ensured that our analysis focused on high-frequency errors, defined as those occurring at least 100 times in the corpus, thereby distinguishing systematic errors from isolated mistakes.

In the next step, following Keshavarz’s error classification catalogue [6], we classified the errors into four categories. Also, in the process of error classification, we defined a sentence as obscure when its meaning was not straightforward or the message was blurred. Additionally, in sentences with multiple errors, each error was counted separately [71]. This classification process was validated through Cohen’s Kappa inter-rater reliability measure, achieving a 98% agreement rate, ensuring consistency and accuracy in our error classification.

Four error classifications identified are as follows: (A) grammatical errors, which included sentences deviating from accepted norms of English grammar, encompassing misplaced and dangling modifiers, conjunctions, cohesion and cohesive devices, omission/misuse of articles/determiners, tenses, run-on and long sentences, and agreement; (B) lexical and semantic errors, which included errors related to word meaning or choice, and orthographic errors such as spelling errors, collocations, and obscure sentences; (C) non-grammatical errors consisted of errors not related to the English language grammar that included punctuation errors in the corpus; and (D) other types of errors with frequencies under 100. Figure 2; Table 1 present an overview of the high-frequency errors in the corpus.

Fig. 2.

High-frequency errors in the learner corpus. Note: Numbers are in percentages

Table 1.

High-frequency types of errors with examples

Types of the error	Definition	Example
1- Agreement	The agreement of a verb with its subject and a pronoun with its antecedent	1- ‘The responsibilities of the ministry of health is that…’ 2- ‘The nurse and doctor takes care of …’
2- Run-on and long sentences	Two complete sentences are squashed together without using a coordinating conjunction or proper punctuation.	1- ‘The government should provide vaccines therefore the cost will be low.’ 2- ‘For example hepatitis B is one of the main challenges in Iran, its mortality rate is high.’
3- Omission/misusing of articles/determiners	Misusing or omitting definite or indefinite articles or determiners	1- ‘… parent thinks that problem of vaccination is…’ 2- ‘Without vaccination the death caused by in the Iran will be high.’
4- Cohesion and cohesive devices	The connectedness of different parts of the text together and is mainly at sentence level	1- ‘… medicine is growing, in addition to it is needed …’ 2- ‘… the governments should pay attention to …’
5- Punctuation	The use of standardized marks, such as periods and commas to clarify meaning	1- ‘On the other hand I think that …’ 2- ‘… it is however seems to…’
6- Collocation and word choice	A predictable combination of words	1- ‘It is not sure to use medication…’ 2- ‘… in respecting for …’
7- Obscure sentences	Not clear in meaning, ambiguous	1- ‘… process in the hospital with the patient is.’ 2- ‘The government the money that should help …’
8- Conjunctions	A word that does not appropriately connect words, phrases, or clauses	1- ‘… parents who they know the benefits of vaccines …’ 2- ‘… although vaccination is important but parents …’
9- Spelling	Deviations from conventionally accepted form of spelling	1- ‘… in our contry, some family do not know the risk of not vaksination …’ 2- ‘Althou the vacine is free in Iran but some parents are afrade …’
10- Tenses	Deviations from grammatically correct uses of tenses	1- ‘Medical equipment have were used in some centers ….’ 2- ‘… vaccination as a preventative tool should be using in earlier ages …’
11- Misplaced and dangling modifiers	A word or phrase that doesn’t modify the correct word or phrase	1- ‘to improve the vaccination, the searching process should be changed.’ 2- ‘in order to reduce child’s stress during vaccines, the use of toys are recommended.’

Note: Definitions of identified errors were adopted from English Grammar in Use (Murphy, 2019)

High-frequency linguistic patterns and the percentage of high-frequency errors in each pattern

In the second part of the Results section, we used data obtained from Wordsmith Tools and sifted manually through the learner corpus to identify high-frequency linguistic patterns and the percentage of high-frequency errors in each pattern. In doing so, we set the minimum frequency for a linguistic pattern to be considered high-frequency at appearing 100 times and more in the corpus. In order to analyze the prevalence and frequency of each linguistic pattern individually, we counted patterns separately when a sentence contains multiple patterns. After confirming 91% of the high-frequency patterns detected in the corpus with three grammar experts, we proceeded to exclude the remaining 9% due to a lack of consensus among our expert panel. The identified high-frequency patterns (Table 2) are discussed below. Also, exemplary sentences including high-frequency linguistic patterns in the learner corpus are available by request.

Table 2.

Identified high-frequency patterns and the percentage of their occurrence in the corpus

High-frequency patterns	%
Central modals	7.59
Informal features	3.62
Hedges	11.42
Boosters	7.35
Discourse markers	22.07
	Total: 52.05%

Central modals

Participants used a sizable number of central modals in their essays. Our finding confirms the results of precursory research [72] that both English native and non-native speakers use almost the same pool of central modals but with different frequency of occurrence. Hyland discusses that this similar pattern may be the result of focusing on teaching modals, which often appear in ESL/EFL textbooks with a high frequency [42]. Table 3 shows the percentage of sentences with high-frequency central modals in the learner corpus.

Table 3.

High-frequency central models in the Corpus

Central modals	%
can	1.39
can not (can’t)	0.96
could	0.87
could not (couldn’t)	0.80
should	1.19
should not (shouldn’t)	0.75
will	1.09
will not (won’t)	0.54
	Total: 7.59%

Informal features

Following Biber et al. [73], we employed a measurement technique involving the summation of normalized frequencies of informal features to quantify the level of informality. Normalized frequencies are defined as a way to adjust raw frequency counts from texts of different lengths so that they can be compared accurately. Normalized frequencies were calculated using a process known as normalization, which involves dividing the raw frequency of each informal feature by the total number of words in the corpus and then multiplying by 1,000 to obtain a standardized rate per 1,000 words [73]. Our investigation focused on five frequently occurring informal features, which also appeared no fewer than 100 times in the corpus (see Table 4). These features were selected from the works of Hyland and Jiang [46] and Chang and Swales [58], and were also commonly referenced in textbooks [74]. Furthermore, these high-frequency informal features were studied in precursory works as representative examples of informal style [75, 76].

Table 4.

High-frequency informal features in the corpus

Informal features	%
1. First person pronouns (I, we, me, us, my, our, mine, and ours) Example: ‘I think that the injection of vaccines’	0.73
2. Unattended anaphoric pronouns (this, that, these, those) Example: ‘This shows that parents in rural areas might not be interested in cooperating with local healthcare to’	0.59
3. Sentence-initial conjunctions or conjunctive adverbs Example: ‘And the fact that vaccines are for free is	0.63
4. Listing expressions such as: and so on, and so forth, etc., used when ending a list Example: ‘ which has some reasons like low economic status, lack of accessibly to healthcare network, lack of knowledge, etc.	0.55
5. Contractions Example: ‘Also, I don’t think that it would help in small cities	1.12
	Total: 3.62%

Hedges and boosters

Various lexico-grammatical markers encompass the expression of hedges and boosters. These markers comprise modal verbs (e.g., may, would, must), lexical verbs (e.g., suggest, think, show), adverbs (e.g., possibly, certainly, obviously), adjectives (e.g., probable, potential, evident), phrases (e.g., in fact, in my opinion), indirect and parenthetical constructions, passives, and if-clauses. Additionally, according to Hyland [34, 39], discourse strategies play a role in hedging and boosting. For instance, addressing encountered difficulties and referencing deficiencies and alternative explanations can be viewed as hedging strategies. Conversely, employing consensual understandings based on shared community membership can be considered as a boosting strategy [39]. Furthermore, Hyland posited that the majority of hedges and boosters consist of lexical items [34].

In the present study, we analyzed the EFL medical corpus for the presence of hedging and boosting elements, using reference lists proposed by Hyland [28]. Additionally, each modifier in the corpus was contextually analyzed by reading the surrounding text, which is a significant step in identifying hedges and boosters since “linguistic forms are complex and the functions they express cannot be identified in a social and textual vacuum” [29]. Also, in the EFL medical corpus, the presence of clusters of hedges or boosters was considered as a singular manifestation of the pattern, thus, treated as one occurrence (e.g., e.g., may suggest, clearly show).

Results show that EFL medical students have a tendency to use more lexical verbs and adverbs (such as: think, suggest, seem, probably, perhaps) than auxiliary modals (such as: may, might) to represent the hedging concept. Also, personal opinions (such as: in my opinion, to some extent) ranked next in terms of frequency. Additionally, another form of hedging, the passive voice, was frequently observed in the present learner corpus (2.48%). The high-frequency of passive voice maybe due to the interference of participants’ L1, as the passive voice is seen predominantly in the Persian writing. On the other hand, phrases (e.g., in fact) and modal verbs (e.g., must) were the most prevalent boosters in the corpus. Tables 5 and 6 review some of the high-frequency hedges and boosters in the corpus, demonstrating that the percentage of high-frequency hedges used by EFL medical students was more than that of boosters.

Table 5.

Hedges in the learner corpus

Hedges	%
passive voice	2.48
think	2.50
suggest	1.29
seem	0.89
perhaps	1.01
probably	0.81
may	0.76
might	0.54
in my opinion	0.60
to some extent	0.54
	Total: 11.42%

Table 6.

Boosters in the learner corpus

Boosters	%
in fact	1.93
must	1.72
always	1.22
actually	1.09
certain/certainly	0.86
clear/clearly	0.53
	Total: 7.35%

Discourse markers

In the present study, Fraser’s taxonomy [35, 37] was adopted, and the dataset provided by Wordsmith Tool was also used to categorize the discourse markers. The markers were then organized into four distinct categories as outlined below:

Results demonstrate that the usage of elaborative markers, which serve to elaborate on previously discussed information [37], was found to be the most prevalent, accounting for 9.16% of the total usage. High-frequency of this category highlights the EFL medical students’ inclination towards providing deeper explanations and elaborations in their writing. This is followed by contrastive markers (5.89%), utilized to bridge opposing segments within a sentence, and inferential markers (4.25%), which convey consequential messages based on preceding information [37]. Temporal discourse markers, on the other hand, were found to be used least frequently, accounting for only 2.77% of the total usage. This suggests a lesser inclination towards expressing temporal relations [36]. Fraser classified temporal discourse markers as a subclass of markers [36]; however, Fraser later excluded this classification [37], with the justification that discourse markers solely reflect semantic relationships between discourse segments. Nevertheless, we posit that discourse markers not only reflect semantic relationships but also serve to display discourse relations. As a result, we include Fraser’s temporal class of discourse markers in our study. Furthermore, each discourse marker is categorized into one of five syntactic categories: coordinate conjunctions, subordinate conjunctions, prepositions, prepositional phrases, and adverbials [36](Table 7).

Table 7.

Four categories of high-frequency discourse markers in the learner corpus

Elaborative	%	Contrastive	%	Inferential	%	Temporal	%
Also	2.01	But	2.61	In conclusion	1.03	Finally	0.73
And	5.52	Although	1.72	Since	0.67	First of all	0.55
For example	1.04	However	1.00	Then	0.54	When	0.95
In addition	0.59	While	0.56	Therefore	2.01	Now	0.54
Total: 9.16%		Total: 5.89%		Total: 4.25%		Total: 2.77%

After analyzing the percentage of high-frequency errors in the five commonly used linguistic patterns identified in the learner corpus, we found that the classification of grammatical errors within the pattern of discourse markers was the most prevalent, accounting for 11.47% of the total errors detected. High-frequency grammatical errors in hedging patterns constituted the second most common category, comprising 3.23% of the total errors. This was followed by errors in central modals (1.96%), boosters (1.41%), and informal features (1.15%). The prevalence of grammatical errors, particularly in discourse markers and hedging patterns, can be attributed to the influence of the participants’ L1, Farsi, which has different syntactic structures and rules compared to English. The transfer of linguistic patterns from the learner’s L1 is a common source of error in their L2 acquisition [5, 6, 66].

Discussion

In the present study, we identified high-frequency errors and linguistic patterns in EFL medical students’ English writing through a learner corpus. We detected 8,691 errors – on average, one error in every two sentences and eight errors in every essay. Also, the data collected showed that the majority of errors were of grammatical nature which is in alignment with precursory studies [14–16]. According to the results, only five out of 11 high-frequency errors in this study, namely cohesion and cohesive devices (16%), omission/misusing of articles/determiners (14%), conjunctions (10%), misplaced and dangling modifiers (9%), and collocation and word choice (12%) account for 61% of the total number of errors EFL medical students made in their writing. We, thus, argue that it is advantageous to prioritize these five prevalent errors in the writing and grammar module within the EMP curriculum in EFL contexts. Specifically, results suggest that emphasis should be placed on teaching cohesion-building skills and the correct use of articles/determiners. Also, obscure (6%) and long sentences (7%) in the category of grammatical errors are of lower frequency compared to other errors in this category.

Lexical errors of collocation and word choice as well as spelling errors account for 15% of total errors. Participants’ lack of proper knowledge regarding formulaic expressions and collocating words is the major cause of the collocation error [6]. In this regard, Hyland and Tse argue that focused analysis of concordance in learner corpora help students learn collocations [77]. Additionally, the spelling error might be attributed to either carelessness, L1 influence, or poor orthographic knowledge [5]. The only non-grammatical error we identified in the corpus is the punctuation error (5%) that may be attributed to students’ lack of L2 punctuation knowledge [64]. In addition, Keshavarz (2008) argues that lack of proper knowledge of discourse markers is a major cause of the cohesion/cohesive error [6].

Meanwhile, L1 influence has a pivotal effect in all classifications of error in L2 writing [78], and in our study particularly resulted in the dangling modifiers, omission of determiners, and long sentences. This influence was identified through a detailed analysis of the error patterns, which closely mirrored the syntactic and grammatical structures of participants’ L1, Farsi. For instance, the flexibility in the placement of modifiers in Farsi often led to dangling modifiers in English. Similarly, the absence of articles in Farsi resulted in the frequent omission of determiners in English sentences. Additionally, the tendency to construct longer, more complex sentences in Farsi contributed to the occurrence of run-on sentences in English.

In a recent error analysis study, Milewski, Fareh, and Al-Sabbah (2022) investigated the relationship between essay genre and writing errors, focusing on whether genre influences the frequency and distribution of error types. Their findings reveal that rhetorical analyses contained twice as many errors as argumentative essays. The study primarily examined error categories and types, such as grammar and lexical errors, and attributed error differences to factors like essay genre, composition method, and submission conditions [79]. While Milewski et al.’s and our study address writing errors in EFL contexts, our research is distinct in its focus on EMP, the high-frequency errors and linguistic patterns, and the practical applications for EMP instruction.

Although high-frequency linguistic patterns in writing depend on disciplinary features and vary across disciplines, results of the present study shed light on the specific usage patterns in EFL medical students’ English writing. Comparing the frequency of discourse markers in L1 and L2, Dumlao and Wilang report that English native speakers use frequently elaborative discourse markers followed by temporal markers [80]. With respect to English non-native speakers’ writing, they claim that inferential discourse markers as well as temporal markers are frequently used. However, EFL medical students’ writing in this study presented a different pattern in which elaborative discourse markers were frequently used, followed by contrastive and inferential markers. On the other hand, temporal markers were used less frequently. Also, Schourup claims that English native speakers frequently use discourse markers in the introductory part of the sentence to help signal the rest of the message whereas non-native speakers use them less frequently at the beginning of a sentence [81]. However, EFL medical students in the present learner corpus used discourse markers mostly positioned at the beginning of sentences which is not consistent with Schourup’s finding.

Results of Hyland’s study on hedges and boosters also show that hedges were used more than boosters in humanities/social sciences [39]. He then argues that writers in humanities/social sciences prefer hedging because they usually use personal judgments and interpretation in their writings; however, writers in fact-based fields (i.e., medical and engineering fields) prefer using more boosters. Contrary to Hyland (1998) and other similar previous research conducted in humanities/social sciences [31, 82], our study with EFL medical students revealed a different pattern. The findings indicated that hedges were utilized more frequently (11.42%) compared to boosters (7.35%) challenging Hyland’s (1998) assertion that fact-based fields like medicine generally favor the use of boosters over hedges. It also suggests that EFL medical students, despite the nature of their field, tend to employ more cautious and tentative language in their academic writing. Additionally, in another similar EFL context, Akbas and Hardman reported that both native English speakers and Turkish speakers of English at the postgraduate level exhibited comparable patterns, preferring to use a greater number of hedges when making claims about their knowledge, in contrast to native Turkish speakers writing in Turkish [83].

Also, the limited presence of informal elements in hard science disciplines, as evidenced in prior studies [46, 58, 84], has been corroborated in this current investigation, accounting for only 3.62% of the total. Nonetheless, the identification of informal features and the reliability of the approach employed - calculating the sum of the occurrences - are primarily reliant on earlier scholarly works [46, 58] needs further validation. Meanwhile, results revealed the high-frequency central modals in the analyzed corpus, illustrating that central modals of ability (4.02%) had the highest prevalence followed by obligation (1.94) and future actions (1.63%) within the corpus. The use of negation was also observed (3.05).

The results highlighted that 19.22% of the total errors observed in the corpus occurred in high-frequency linguistic patterns. Among these, grammatical errors were more prevalent than other classifications of errors. Specifically, the analysis revealed that the majority of errors in high-frequency patterns were identified in sentences containing discourse markers (11.47%) and hedges (3.23%), indicating that these two linguistic patterns are more prone to grammatical errors compared to other linguistic patterns.

The high occurrence of grammatical errors, especially in discourse markers and hedging patterns, can be linked to the transfer of linguistic patterns from L1, Farsi, which has different syntactic structures compared to English. Additionally, the complexity of discourse markers and hedging requires a nuanced understanding of context and usage, which can be challenging for EFL learners. This complexity, combined with the syntactic differences between Farsi and English, likely contributes to the higher frequency of grammatical errors in these patterns.

Limitations and future research

This study has some limitations that warrant attention. The varying lengths of the essays might have impacted the frequency of modifier usage by individual students, which should be taken into account during results analysis. Additionally, it is plausible that some essays in this study contain alternative means of hedging or boosting statements that may have been overlooked. Therefore, the modifiers examined here may offer a broad overview of a phenomenon that necessitates further investigation.

Future research could address these limitations by employing a more standardized essay length to ensure consistency in frequency analysis. Additionally, a more comprehensive examination of hedging and boosting strategies, including those not covered in this study, would provide a deeper understanding of these linguistic features. Future studies could also incorporate a longitudinal approach to observe changes in modifier usage over time and across different proficiency levels. Finally, expanding the corpus to include a more diverse range of medical students from various linguistic backgrounds could enhance the generalizability of the findings.

Conclusion

We argue that findings of the present study could contribute to the improvement of EFL medical students’ writing as well as EMP curriculum at medical universities in EFL contexts. In particular, as the errors are highlighted, the results may lead to consciousness raising in students that would ultimately lead to a faster improvement of their English writing. Meanwhile, the findings of our study offer insights that highlight the specific high-frequency errors that should be prioritized in classroom remedial activities, especially with medical students in EFL contexts. A further implication of this study is its prognostic power, as the findings of this study might help curriculum developers and English language faculty members at medical universities in our context and other similar EFL contexts revise their EMP curriculum and course plans with a focus on high-frequency errors and patterns identified in the present study.

Appendix 1

Writing Activity.

Topic

Should vaccinations be mandatory for all children?

Prompt

Vaccinations have been a topic of controversy for many years. While some argue that they are necessary to prevent diseases and protect public health, others argue that they can be harmful and that parents should have the right to choose whether or not to vaccinate their children. Write an essay no less than 250 words in which you take a stance on whether or not vaccinations should be mandatory for all children.

Task requirements:

• Use evidence to support your argument.

• Address potential counterarguments and explain why your stance is the most reasonable.

Author contributions

Both authors (S.J.E. and A.D.) conducted all steps of the study together.

Funding

There is no source of funding for the present research.

Data availability

The datasets analyzed in this study are available upon reasonable request from the corresponding author.

Declarations

Ethics approval and consent to participate

This study was approved by the Ethics Committee of Iran University of Medical Sciences (IR.IUMS.REC.1403.120). The written informed consent was obtained from all participating students.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.