1. Intrduction

The latest edition of the world's largest ranking of countries by English skills, carried out by EF EPI (^{Education First English Proficiency Index, 2021}), indicates that out of 112 countries, Costa Rica ranks 44 and has moderate English proficiency with a score of 553 (vis-à-vis the Netherlands, which has very high proficiency and ranks 1st on the list with a score of 663). Some may argue that a survey of this type cannot by any stretch paint a completely accurate picture due to sampling procedures⁽¹⁾, yet it certainly shows a preview of a larger reality: that the English proficiency level of Costa Rican pupils and youngsters seems to be stagnant. In fact, recent news reports pose a problem that authorities of the Ministry of Public Education (MEP in Spanish) have yet to grapple with. In 2021, the Foreign Language Assessment and Training Program (PELEX in Spanish) from the School of Modern Languages of the University of Costa Rica administered a language competence test in all public high schools nationwide. The results were not encouraging: 64% of the students were placed on the A1 or A2 band based on the Common European Framework of Reference for Languages. Such results did not yield the B1 minimum that MEP was hoping for, and they certainly do not look promising to reach bilingualism by 2040 (^{Ruiz, 2022}).

The foregoing implies that teaching English could represent a daily challenge in the life of Costa Rican second language (L2) practitioners generally and L2 writing teachers particularly, especially considering that any L2 issue that highschoolers may drag could show at the university level. Hence, this scenario calls for one action that could be useful in furthering knowledge of English teaching in a context with a clear educational need: learner corpus research. This line of inquiry “has primarily relied on collecting and analyzing second language … learner writings” (Granger, 2008 cited in ^{Alexopoulou et al., 2017}, p. 1) with the purpose of, among other things, identifying frequency of use of given L2 structures (^{Neff et al., 2004}) and ascertaining areas of L2 struggle (^{Arjan et al., 2013}). Indeed, the fact that learners' language collections can be computerized has made it possible to have large learner corpora of over 40 million words (e.g., the Cambridge Learner Corpus) as well smaller ones collected with a specific research purpose in mind (^{Díaz-Negrillo, 2009}).

However, there is narrowed down corpus data about the linguistic problems of EFL learners from various first language (L1) Spanish backgrounds: available knowledge emerges mainly from EFL learners in Spain—be it from large (^{Díaz-Negrillo and Valera, 2010}) or small learner corpora (^{Díez-Bedmar, 2005}). What is more, to the best of the researcher's knowledge, no major college wide study has been conducted in the context of this investigation. Hence, in an attempt to assist in the understanding of EFL in Costa Rica generally and from an undergraduate standpoint specifically, the present corpus-aided investigation seeks to identify the L2 error patterns of Costa Rican university writers at all academic levels of an English major. Specifically, the research question that guided this study was the following: What are the grammatical and non-grammatical L2 error patterns of university writers across academic levels of an English major of a public university?

2. Theoretical background

With the advent of Contrastive Analysis (CA) in the late 1950s and Error Analysis (EA) in the 1960s, researchers sought to analyze learners' L2 errors by looking for differences between learners' L2 and first language (L1) (i.e., CA) and to classify L2 learners' errors to explain what caused them (i.e., EA) (for a review, see ^{Bitchener and Ferris, 2012}). From these studies (^{Bhela, 1999}), it was possible to determine that learners' L1 may have an influence on L2 written inaccuracies. Specifically related to Spanish L1, different researchers have shed light on the nature of errors of speakers learning English as a FL. One such example is ^{Alonso (1997)}, who conducted a study with twenty-eight first year EFL high school students in Spain. According to the author, errors from compositions about the last film the participants had seen were mostly interlingual errors, that is, those “that reflect the learner's first language structures” (^{Dulay et al., 1982}, p. 23). Similar to ^{Alonso (1997)}, Calsín (2011, cited in ^{Vargaya, 2019}) analyzed the participants' texts—in this case, 4th and 5th year Linguistics and English students—and found Spanish L1 influence on written errors related to the absence of the –s for the third person conjugation in simple present tense (omission error), the unnecessary addition of –s in adjectives (addition error), and the lack of accuracy in placing the adverbs of frequency in the correct order (lack of sentence order).

Nevertheless, criticism to EA and CA theories because they were too limited in their focus (^{Bitchener, 2016}; ^{Ellis, 1994}), on the one hand, and the incorporation of computers in data collection, on the other hand, shifted empirical efforts to a line of inquiry with a methodology that studies language use beyond the causes of L2 errors and L1 comparisons to understand them: that is, corpus linguistics. ^{Lindquist (2009)} defines corpus as “a collection of texts which is stored on some kind of digital medium and used by linguists to retrieve linguistic items for research or by lexicographers for dictionary-making” (p. 3). As a result, there are large native corpora that contain all sort of samples of English, which is the most studied language thus far (^{Granger, 1998a}). Some of these are the Brown/Frown Corpus, the London-Lund Corpus of Spoken English (LLC), the Bank of English (BoE), the British National Corpus (BNC), the Corpus of Contemporary American English (COCA), and the International Corpus of English (ICE). Interestingly, the emergence of native English corpora made it clear that there was also a need for corpora that studied English as used by L2 learners, hence the term learner corpora (^{Díaz-Negrillo, 2009}; ^{Nesselhauf, 2004}). Among the most prominent learner corpora are the International Corpus of Learner English (ICLE), the Longman Learners' Corpus (LLC), and the Hong Kong University of Science and Technology (HKUST) learner Corpus (for a comprehensive list, see ^{Lindquist, 2009}; ^{Pravec, 2002}). Currently, learner corpora “give us access not only to errors but also to learners' total interlanguage” (^{Granger, 1998b}, p. 6)⁽²⁾. One lack, however, is that much of the understanding of English errors at a university level comes from seminal work on native-speaker corpora (Connors and Lunsford, 1988; ^{Hodges, 1941}; ^{Johnson, 1917}; ^{Lunsford and Lunsford, 2008}; ^{Witty and Green, 1930)}, and when studies with EFL university learners have been conducted, the context is situated mainly in Europe (e.g., ^{Dagneaux et al., 1998}) and Asia to a lesser extent (^{Narita, 2013}).

Thus, few of the investigations on the overall written production of Spanish L1 EFL university writers are Díaz-Negrillo and Valera (²⁰¹⁰), ^{Neff et al. (2004)}, and Díez-Bedmar (²⁰⁰⁵), out of which just two explore learners' errors. To illustrate, Neff et al. investigated fourth-year university learners' lexico-grammatical patterns of writer stance (e.g., it is + (adverb) adjective + that; it is + (adverb) said/thought + that) and compared them with those of professional writers and English L1 university students. The participants were EFL writers whose first languages were Dutch, Belgian-French, Italian, and peninsular Spanish and the language samples were extracted from ICLE. Main results showed an overuse of it is + (adverb) adjective + that and the agentless passive by the EFL learners, whereas the it is + adjective pattern showed no significant differences. Different from Neff et al., Díez-Bedmar (²⁰⁰⁵) analyzed first-year students' essays to identify L2 learners' written errors at a morphological, syntactic, semantic, and pragmatic level. Overall, the findings revealed that some of the most problematic areas were punctuation, spelling conventions, verb tenses, and articles. Then, as an error frequency study, Díaz-Negrillo and Valera (²⁰¹⁰) examined a sample of the Non-native Corpus of English (NOCE, ^{Díaz-Negrillo, 2009}) and found a complex picture where comma usage, for example, seemed highly problematic along with lexical issues such as wrong word choice.

Clearly, albeit their significant findings, previous investigations are not enough to gain sound insight into Spanish L1 EFL learners' interlanguage and to inform in turn L2 educators and researchers alike. Consequently, the need to further broaden current knowledge of Spanish L1 EFL university writers at different academic levels in Costa Rica inspired this study.

3. Methodology

4. Results and Discussion

The research question that guided this study sought to identify the grammatical and non-grammatical patterns of university writers at all academic levels of an English major of a public university. Table 2 displays the descriptive statistics of ranked error patterns in first-year university writers. Table 3 presents the descriptive statistics of ranked error patterns in second-year university writers. Table 4 summarizes the descriptive statistics of ranked error patterns in third-year university writers. Table 5 shows the descriptive statistics of ranked error patterns in fourth-year university writers.

As can be seen, the areas of linguistic issues differed across academic levels. Even though first-year learners' number one error category was lexical derivation (n = 71), overall, the error categories with higher occurrences were grammar oriented, ranging from subject-verb agreement issues (n = 68) and fragment (n = 64) to word order (n = 52) and (in)definite article confusion along with subject deletion (n = 49). A few non-grammatical issues also appeared at the top, namely comma splice (n = 70) followed by comma omission before coordinating/correlative conjunction joining clauses (n = 53).

It can also be observed that when compared to their first-year counterparts (cf. Table 2), some grammatical error categories remained in the top 10 of second-year writers. Table 3 reveals that such is the case of sentence fragment (n = 67) and missing or overinclusion of definite article (n = 60) errors. A similar situation occurred with a non-grammatical error type such as spelling, which ranked high both in first (n = 46) and second year (n = 54). In addition, punctuation-related issues that ranked lower in first year (e.g., unnecessary comma before coordinating/correlative conjunction joining words or phrases) had a higher ranking in second year (n = 62). Others remained equally problematic, for example, comma splices (n = 62) and comma omission before a coordinating/correlative conjunction joining clauses (n = 55). As far as lexical errors are concerned, word formation (n = 56) and word choice (n = 51) issues had lower counts unlike missing lexical items, which increased (n = 42).

From Table 4 it can be seen that third-year writers' predominant error categories consist of non-grammatical issues, being comma-related errors in the top three. It is also evident that out of the six types of lexical errors, lexis.derivation (n = 40) lexis.misselection (n =39)—albeit the lower frequency of occurrence when compared with those of first- and second-year learners—were still troublesome. Similarly, despite the lower sum, grammatical categories such as parallelism (n = 37) word order (n = 37), sentence fragment (n = 29), and article-related issues were ranked high. It is also worth highlighting that error categories involving an unnecessary comma between verb and object as well as comparative adjective issues had no error counts, which was not the case in first (cf. Table 2) and second (cf. Table 3) year.

Table 5 shows a similar pattern to Table 4: (a) error types with a higher frequency of occurrence were non-grammatical rather than grammatical, (b) lexis derivation remained in the top five, and (c) the grammatical issues in both academic levels (i.e., 3rd and 4th year) were the same except that they had lower counts (i.e., parallelism, sentence fragment, and word order issues). One difference, however, is that there was no error trace in seven error categories, out of which three were non-grammatical and three were grammatical.

Thus far, Table 1 to 5 clearly render an intricate linguistic scenario. That is why the theoretical and practical implications emerging from the results will be explained in light of key methodological variables from previous research (4.1) as well as relevant factors in the EFL classroom (4.2).

5. Conclusion

The present corpus-aided study widens current knowledge of the error patterns of L2 university writers generally and Costa Rican EFL learners at UCR in IIC2019 particularly. In a nutshell, main findings rendered a complex linguistic scenario worth highlighting: (1) even if first-year learners' highest error frequency was lexically related, the predominant L2 issues were more grammar oriented, (2) second- and first-year learners had similar grammatical issues on top, yet punctuation issues in comma usage started to rank higher as learners progressed in the major, and (3) while some error frequencies decreased in fourth year to the point of not appearing at all, some lexical and syntactic matters—at a phrase and clause level—remained problematic across academic years. These findings lend support to the belief that if there is something that “corpus research has helped clarify is error” (^{Wilder and Yagelski, 2018}, p. 384).

In fact, even though the present results emerge from a particular L2 learning environment, this study could still be illuminating for stakeholders in similar circumstances. First, it might be common belief that once L2 learners pass a course and advance in their study plan, they should show L2 improvement (even mastery) of the L2 linguistic content they were exposed to. Nevertheless, as a contribution to L2 education, the results refute this common misconception and show that this may not always be the case and that, as far as writing is concerned, L2 error frequencies could vary across and within academic levels. Such findings have relevant pedagogical and theoretical implications because they add support to SLA research, which has stated that L2 acquisition is complex, dynamic, and non-linear (^{Larsen-Freeman, 1997}, ²⁰⁰³). Therefore, if L2 exposure does not immediately equate with L2 development—let alone L2 acquisition (for a distinction, see Bitchener and Storch, 2016, p. 2), L2 practitioners may want to reflect on their teaching practices. For instance, keeping in mind the Theoretical Framework for the L2 Learning Process in SLA (see ^{Leow, 2015}, p. 15) and the Stages of Cognitive Processing of Input (see ^{Bitchener and Storch, 2016}, p. 18), a good starting point would be asking oneself: How many classroom activities am I implementing to maximize learners' chances to consolidate L2 knowledge? Am I providing written CF? Am I making sure learners process such feedback? Am I exposing learners to sufficient TL input?

Interestingly, in the context of this investigation, the latter may also be worth asking to interested parties at the highest levels of government (e.g., MEP authorities) since at primary and secondary levels, teaching English in Spanish has been customary (^{Ugarte, 2015}), which clearly is a detrimental practice for learners' L2 development and the country's goal to reach bilingualism by 2040. Hence, in non-predominant English countries, supervision and training from decision-making authorities are also needed for any real L2 change to be seen nationwide. Clearly, seeking to raise awareness and understanding of how a L2 is learned is necessary not only for those at the front line of the L2 classrooms but also for those on top at the ministry level. In this respect, the present study offers a springboard for such a discussion, contributing in turn to Costa Rican L2 education.

Second, the results of this study could also be of use for corpus linguistics researchers to inform their research design and account for all variables. To list one example, the present results differed from those of other studies on error identification (^{Amiri and Puteh, 2017}) because of differences in key variables (e.g., task type). This was somewhat expected because as ^{Granger (1998a)} states, “learner output has been shown to vary according to the task type” (p. 8). As a matter of fact, the author further adds that “the topic is also a relevant factor because it affects lexical choice, while the degree of technicality affects both the lexis and the grammar” (p. 8). Nevertheless, while not surprising, differences in findings do bring to the fore the need to rigorously report all variables to determine to what extent research findings may (or not) be applicable to other learning environments. For this same reason, caution should be exercised in the interpretation of the results in this study.

Third, it is hoped that L2 practitioners and L2 learners alike can benefit from the bird's eye view of the L2 error patterns of the EFL learners in this investigation. That is, the fact that the error frequencies of some L2 error categories still ranked high over time seems to suggest that learners' L2 knowledge of lexical, syntactic, morphological, and stylistic domains could need more expert input (in the form of explicit instruction and/or feedback) depending on the complexity of the target structure—an aspect already touched upon in the bulk of written CF studies (^{Bonilla-López et al., 2021}; ^{Diab, 2015}). Taking this into consideration, the present findings may be useful to increase L2 practitioners' awareness of potential areas of struggle of FL college learners and to inform, as a result, the creation of classroom materials that will cater to their students' linguistic needs.

Finally, future studies might want to consider the following caveats. It is hard to characterize an entire learner type on the basis of a small learner corpus based on one text type and collected at a given point in time (i.e., synchronic). Therefore, studies that aim for a larger sample and that create a learner corpus consisting of varied rhetorical patterns, emerging from one prompt only (per pattern), and having similar text length remain in order. Doing so would improve control of variables such as task, topic, and text length and allow a fairer comparison among learners (see ^{Caines and Buttery, 2018} for an explanation of opportunity of use). As a matter of fact, in the context of this investigation, there is a clear need to conduct a nationwide corpus investigation that gathers a variety of texts both at a high school and university level and for larger stretches of time (i.e., diachronic). In other words, besides administering much-needed L2 competence tests (e.g., PELEX efforts), analyzing learners' actual L2 output through corpus data may be the only way of painting a complete picture of the country's L2 English situation as far as proficiency is concerned. In this respect, while targeted at college level, the present corpus-aided investigation constitutes a start in that direction.