Writing Assessment and Linguistic Patterns in University Student Narratives

Abstract

This study examines the relationship between writing performance, assessed with the Early Writing Alert System (SISAT), and linguistic patterns in student narratives from one public and one private university in northeastern Mexico. Variables such as lexical density and richness, text volume, and thematic progression were analyzed to explore how institutional context influences narrative writing and its assessment. A non-experimental, descriptive–comparative design with interpretive triangulation was employed. The corpus comprised 148 narratives produced over three academic periods, analyzed using automated linguistic tools alongside SISAT scores. Descriptive statistics, Spearman correlations, and Kruskal–Wallis tests were applied to examine differences between the two institutions and across periods. The results indicate intermediate performance at both universities, with differentiated patterns: at the public university, lexical richness and density positively correlated with SISAT scores, while greater text volume was negatively associated; at the private university, both text length and diversity were positively related, though excessive lexical density appeared counterproductive. No statistically significant differences were observed between periods or between the two universities. Our findings highlight that quantitative linguistic indicators complement normative assessment and underscore the role of institutional context in writing development. The study also emphasizes the formative and expressive functions of narrative writing, supporting pedagogical strategies that integrate automated assessment with qualitative analysis to foster self-regulation, symbolic expression, and ethical reflection.

1. Introduction

1.1. Academic Writing in Higher Education: An Integrative Approach

Academic writing in higher education requires an integrative theoretical framework that captures its multiple dimensions. It can be understood simultaneously as a cognitive process, a discursive and rhetorical practice, a linguistic phenomenon, and a situated social activity. Within this perspective, the present study brings together five complementary approaches—rhetorical, linguistic, sociocultural, didactic, and technological—which underpin the automated analysis of university students’ narrative writing and its relationship with institutional writing assessment.

1.2. Writing as a Cognitive and Self-Regulated Process

From the perspective of cognitive psychology, writing has been conceptualized as a complex process involving planning, textualization, and revision, mediated by working memory and prior knowledge (Flower & Hayes, 1981). This model established writing as a recursive and non-linear activity in which writers continually make decisions about content, organization, and audience.

Self-regulated learning theory subsequently expanded this view by emphasizing learners’ active role in planning, monitoring, and evaluating their own writing processes and outcomes (Zimmerman, 2000). Academic writing therefore requires metacognitive skills that allow students to evaluate idea clarity, adjust discourse strategies, and respond to external quality criteria. Empirical studies consistently show that higher levels of self-regulation are associated with more coherent, structured, and genre-appropriate academic texts (Chaverra Fernández et al., 2022; Sandoval-Cárcamo et al., 2024). This perspective is particularly relevant for automated writing assessment, as artificial intelligence-based systems can scaffold self-regulation through immediate, objective, and iterative feedback, thereby supporting cycles of continuous writing improvement.

1.3. Rhetorical and Discursive Perspectives on Academic Writing

From a rhetorical standpoint, academic writing is conceived as a communicative and persuasive act oriented toward a specific audience and governed by shared disciplinary conventions. Its foundations can be traced to Aristotle’s classical rhetoric, particularly the principles of ethos, pathos, and logos, which remain relevant in contemporary academic discourse.

In research on academic genres, Swales (1990) introduced the notion of discourse communities and the CARS (Create a Research Space) model, highlighting the strategic rhetorical moves that structure research writing. Hyland (2002, 2005) further developed this perspective by examining the interpersonal dimension of academic discourse and the role of metadiscourse, stance, and evaluative resources in constructing authorial voice and credibility. From this view, academic writing involves positioning knowledge claims in relation to prior research, anticipating readers’ expectations, and guiding interpretation. Automated narrative assessment can operationalize these principles by identifying discursive organization patterns, thematic development, and argumentative markers relevant to writing quality.

1.4. Linguistic Approach to Academic Writing

The linguistic approach focuses on the formal language features that support meaning construction in academic texts. Halliday and Hasan’s (1976) concept of textual cohesion highlights mechanisms such as reference, ellipsis, conjunction, and lexical repetition as essential for coherence. Within systemic functional linguistics, language is further described through three metafunctions: ideational, interpersonal, and textual (Halliday, 1985).

Research indicates that academic writing typically exhibits high lexical density, frequent nominalizations, abstract nouns, impersonal constructions, and logical connectors, with variation across genres, disciplines, and educational levels (Chyzhykova, 2024; Dong et al., 2023; Lipková, 2024; Istiqomah & Basthomi, 2024; Nguyen & Edwards, 2015). Corpus-based approaches have enabled the empirical description of these features through large-scale analyses of expert and student texts (Biber et al., 1999; McEnery & Hardie, 2012). Learner corpora, in particular, have been widely used for diagnostic, evaluative, and pedagogical purposes in higher education writing research (Boulton & Cobb, 2017; Ma et al., 2023; Ueno & Takeuchi, 2023; Granger, 2024). In this context, automated linguistic analysis allows for the objective measurement of variables such as lexical density, lexical diversity, and grammatical pattern frequency, providing quantitative evidence to support writing assessment.

1.5. Sociocultural Perspectives and Academic Literacy

From a sociocultural perspective, academic writing is understood as a situated social practice shaped by institutional norms, power relations, and cultural values (Barton & Hamilton, 1998; Street, 1984, 2015). Learning to write at university entails appropriating disciplinary genres and legitimized ways of constructing and communicating knowledge (Lea & Street, 1998). This perspective has highlighted the challenges faced by first-generation students, learners from underrepresented backgrounds, and second-language writers, for whom academic writing may function as a mechanism of exclusion (Lillis, 2001).

In response, academic literacy models have emphasized the integration of tutoring, collaborative writing, and formative feedback within disciplinary teaching (Calvo et al., 2020; Davis et al., 2025; Fuster-Barcelo et al., 2025; Zheldibayeva, 2025). Automated assessment systems, by offering continuous and accessible feedback, have the potential to support more equitable writing development, provided they are implemented ethically and in coordination with instructor guidance.

1.6. Didactic Approach and Artificial Intelligence in Writing Instruction

The didactic approach views academic writing as a competence that must be taught explicitly, progressively, and in close alignment with disciplinary curricula. Carlino (2005, 2013, 2023) argues that writing is not a generic skill acquired prior to university study, but a situated disciplinary practice that requires sustained pedagogical support. Consequently, effective instruction should integrate critical reading, guided text production, revision processes, and metacognitive reflection. Narrative and creative writing have also been incorporated as pedagogical strategies to enhance motivation, expressiveness, and engagement with academic discourse (Barbara et al., 2024; Bruner, 2001).

However, the assessment of academic writing remains constrained by subjectivity, inter-rater variability, and instructor workload (Hyland, 2019; Sánchez-Rivas et al., 2023). In this context, artificial intelligence-based language models such as ChatGPT 5.2 offer new possibilities for automated text analysis, enabling the evaluation of coherence, grammatical accuracy, discursive structure, and linguistic patterns, as well as the generation of immediate and adaptive feedback (Amabile, 2018; OpenAI, 2023). Recent studies suggest that these tools support idea organization, linguistic revision, and autonomous learning among both students and pre-service teachers (Güler et al., 2025; Kaur & Kapoor, 2025; Ravšelj et al., 2025). From rhetorical, linguistic, sociocultural, and didactic perspectives, artificial intelligence is therefore conceptualized as a complementary resource that enhances formative assessment rather than replacing teacher judgment.

1.7. Theoretical Articulation and Projection of the Study

Based on this theoretical convergence, this article proposes a technical–methodological model for the automated evaluation of narrative writing based on the architecture of language models such as ChatGPT. The model integrates contributions from rhetoric, linguistics, socioculturality, and didactics, together with the pedagogical benefits of narrative and creative writing. It is proposed that combining traditional pedagogical approaches with artificial intelligence technologies can improve the quality of academic writing, strengthen student self-regulation, and offer immediate and personalized feedback in diverse university contexts. The study aims to analyze the relationship between institutional writing assessment and linguistic patterns present in the narratives of higher education students, identify linguistic and rhetorical indicators associated with different levels of performance, compare institutional contexts, and explore the pedagogical potential of automated assessment as a complement to traditional academic writing assessment.

2. Materials and Methods

Large-scale assessments conducted by the Secretaría de Educación Pública (SEP, Mexico’s Ministry of Education) and the Comisión Nacional para la Mejora Continua de la Educación (MEJOREDU, National Commission for the Continuous Improvement of Education) have consistently reported persistent weaknesses in students’ writing competence, particularly in orthography, textual cohesion, and communicative clarity at the secondary level. Regional assessments such as APRENDE Tamaulipas, a state-level educational assessment program, further indicate that these difficulties affect both public and private institutions, with no sustained improvement over time. However, while these evaluations provide valuable diagnostic information for basic education, little is known about whether such deficiencies persist in higher education, or how they are manifested in university-level writing across different institutional contexts.

2.1. Contextualization of the Study

This study was conducted at two universities located in Tamaulipas, a state in northeastern Mexico. One institution is public, and the other is private. These universities were selected intentionally to explore potential differences in students’ writing performance associated with institutional type. Public and private universities in Tamaulipas differ in terms of student population, available resources, pedagogical practices, and access to academic support, which may influence the quality of written expression. By examining both types of institutions, the study aims to identify not only general patterns of writing competence among university students but also the ways in which institutional context shapes linguistic and discursive performance. This regional focus also contributes to understanding writing development in northeastern Mexico, a context for which there is limited empirical evidence at the higher education level.

2.2. Theoretical Linkage and Operationalization

The study integrates cognitive, self-regulation, rhetorical–discursive, and sociocultural frameworks. These perspectives guided the following: Selection of automatic metrics—lexical density, lexical richness, syntactic complexity, and thematic progression. Narrative analysis—cohesion, planning, and argumentation. Institutional comparison—public vs. private universities to capture sociocultural differences (Table 1).

Table 1. Theoretical–metric linkage.

Theoretical Perspective	Metric/Indicator	Correspondence with SISAT
Cognitive/self-regulation	Lexical density, syntactic complexity	Planning and clarity
Rhetorical/discursive	Thematic progression, connectors	Cohesion and argumentation
Systemic functional linguistics	Lexical richness	Vocabulary sophistication
Sociocultural	Institutional comparison	Performance differences by context
Didactic/technological	AI-based feedback	Complements institutional assessment

2.3. Corpus and Unit of Analysis

The corpus consisted of 148 narratives produced by university students, distributed equally between public and private universities (74 texts per institution) in the first, second, and third periods of 2025. The unit of analysis was the complete narrative text, which allowed for the identification of collective linguistic and discursive patterns. The texts were regular academic productions, without experimental intervention. All participating students provided written informed consent, agreeing to the collection and analysis of their narrative texts anonymously and voluntarily. Confidentiality and the right of participants to withdraw at any time were guaranteed. To protect the identity of the participants, alphanumeric codes were assigned to each text (e.g., S01 and S02).

2.4. Procedure and Analysis of the Texts

The design of the analysis puts into practice the theoretical frameworks presented in the Introduction; the steps for identifying lexical–discursive patterns reflect the integration of cognitive, rhetorical, linguistic, sociocultural, and didactic perspectives in the automated evaluation of academic writing. Thus, each methodological stage was selected and applied to capture both the linguistic competence and the discursive and pedagogical dimensions of the students’ texts.

2.4.1. Institutional Assessment: SISAT

To contextualize the linguistic results, the Early Warning System in Writing (SISAT) developed by the Ministry of Public Education as used. This instrument assesses six writing indicators on a scale from 1 to 3, giving a total score between 6 and 18 and classifying performance into three levels: Needs support (≤9), Developing (10–14), and Expected level (15–18). This assessment provides the institutional benchmark with which the automatically extracted linguistic indicators are correlated. Table A1 presents the SISAT performance scale. A detailed description of the instrument is provided in Appendix A. Table A2 and Table A3 and Appendix B. Student Narratives.

2.4.2. Preprocessing: Tokenization and Lexical Normalization

Before analysis, the texts underwent tokenization and lexical normalization.

Tokenization segments the text into minimal units of analysis—words and punctuation marks—facilitating the identification of lexical patterns, spelling errors, and grammatical relationships. For example, the text “The girl wrote a story.” is tokenized as: (“The,” “girl,” “wrote,” “a,” “story,” “.”). This allows the software to analyze each word independently and calculate automatic metrics such as word frequency, sentence length, and grammatical patterns.

Lexical normalization unifies morphological forms (run, ran, running → run), standardizes the use of lowercase letters, and corrects frequent typos. This step ensures that morphological variation or minor errors do not distort the results, improving the accuracy of metrics such as lexical diversity and syntactic complexity.

2.4.3. Extraction of Linguistic Metrics

After preprocessing, the following metrics were calculated automatically:

(a)

Lexical density and lexical richness, to assess the variety and sophistication of vocabulary.

(b)

Frequency of grammatical categories, such as verbs, nouns, and adjectives, to examine discourse structure.

(c)

Number of sentences and syntactic complexity, reflecting the organization of ideas.

(d)

Thematic progression, to identify content coherence and sequencing.

These metrics allow us to relate objective linguistic variables to the performance levels established by SISAT and to possible differences between public and private universities, in line with cognitive and self-regulated learning frameworks.

2.4.4. Identification of Lexical–Discursive Patterns

Based on the metrics extracted, a qualitative and quantitative analysis of the texts was performed to identify recurring lexical–discursive patterns associated with performance levels and institutional differences:

(a)

Automated analysis: Co-occurrence of words, frequency of grammatical structures, and syntactic patterns.

(b)

Interpretative analysis: Thematic progression, textual cohesion, and discursive organization.

The integration of these techniques allowed for a contextualized diagnosis of written competence, connecting linguistic evidence with institutional assessments and providing a solid basis for pedagogical recommendations. The choice of metrics and procedures was based on the following:

(a)

Cognitive and self-regulation perspectives: To capture planning, clarity, and textual cohesion.

(b)

Rhetorical and discursive principles: To analyze narrative organization, thematic progression, and argumentative markers.

(c)

Systemic functional linguistics: To select indicators of lexical density, richness, and syntactic complexity.

(d)

Sociocultural perspective: To justify the comparison between public and private universities.

(e)

Didactic and technological perspectives: To integrate artificial intelligence-based tools that provide objective, immediate, and individualized feedback, complementing traditional methods of institutional assessment.

2.4.5. Software and Analytical Tools

To ensure a systematic, replicable, and context-sensitive analysis of students’ narrative texts, this study employed a combination of corpus-based analytical frameworks and digital tools specifically selected to support each stage of the linguistic and discourse analysis. The software and tools were aligned with established practices in corpus linguistics and automated writing assessment, enabling the integration of quantitative metric extraction and qualitative pattern identification.

Corpus-Based Analytical Framework

The overall analytical design was based on the principles of corpus linguistics, which allow for the empirical examination of authentic texts using computational methods (Biber et al., 1999; McEnery & Hardie, 2012). This framework guided the selection of linguistic variables, the organization of the textual dataset, and the interpretation of the structural and frequency-based patterns observed in the students’ narratives.

Analytical Tools Developed by the Researchers

Customized tools integrated by the researchers were used for text preprocessing and automated linguistic analysis. These tools enabled tokenization, lexical normalization, metric calculation, and the identification of recurring lexical–discursive patterns. Their design responded to the specific linguistic characteristics of academic writing in Spanish and to the evaluative criteria established by the institutional evaluation system (SISAT), which facilitated alignment between linguistic metrics and educational performance indicators.

Integration and Documentation of the Analytical Procedures

The combination of corpus-based software and researcher-developed tools allowed for efficient processing of the text corpus and ensured consistency across the different analytical stages. Detailed descriptions of these tools, their functions, and access links are provided (see Figure A1, Figure A2 and Figure A3; Appendix C) to ensure transparency and replicability, while avoiding interruptions to the methodological flow of the main text.

2.4.6. Data Analysis

Data analysis followed a multilevel strategy aligned with the study’s objectives:

• Descriptive analysis: Means, percentages, and distributions by academic period and university, providing an overview of the characteristics of the corpus.
• Correlational analysis: Spearman’s correlation to explore relationships between institutional performance (SISAT) and linguistic variables. This was calculated separately for public and private universities, considering institutional differences in resources, student profiles, and teaching practices.
• Comparative analysis: Kruskal–Wallis tests to examine differences between academic periods and between institutions. Although inferential tests are not the main focus, they provide additional context to the descriptive and correlational findings.
• Quantitative–qualitative triangulation: Statistical results combined with analysis of lexical–discursive and rhetorical patterns, offering an integrated interpretation of writing development according to institutional context.

This strategy allows us to link metrics of lexical richness, cohesion, and narrative structure with institutional performance, demonstrating how differences in context can influence writing competence.

3. Results

3.1. Characteristics of the Public University Corpus

The corpus consisted of 74 narratives from public university students, distributed across three academic periods. Lexical density varied between 21.5% and 29.6%, indicating texts of intermediate lexical complexity. Lexical richness (unique words) was higher in the first period and slightly lower in the third period, reflecting the evolution of writing towards more experiential and descriptive narratives (Table 2).

Table 2. Relationship between lexical metrics and writing performance (SISAT) by academic term at public university.

Period	Number of Texts	Tokens/Words	Unique Words	Lexical Density (%)	Average SISAT Score
First	24	27,486	5912	21.5	13.6
Second	25	19,483	5762	29.6	13.7
Third	25	21,584	4872	22.0	12.5

3.2. Linguistic and Thematic Patterns

Nouns related to childhood, family, and school predominate, reflecting an affective and educational focus in the narratives. Four recurring thematic nuclei were identified: family/affection, mystery/fantasy, space/fear, and loss/rebirth. Youth narratives are characterized by tripartite structures (beginning–climax–ending) and limited use of connectors, especially at ‘Developing’ performance levels (Table 3).

Table 3. Relationship between SISAT levels and dominant linguistic features.

SISAT Level	Dominant Linguistic Features	Observations
Requires support	Weak cohesion, limited lexicon, fragmented narration	Low lexical density (<20%), abrupt endings
Developing	Basic cohesion, concrete vocabulary, simple connectors	Medium lexical density (≈21–29%), still standard tripartite structure
Expected level	Clear cohesion, varied vocabulary, functional connectors	High lexical density, complex thematic sentences, reflective closure

Most narratives are at the “Developing” level, indicating functional narrative competence but with room for improvement in cohesion and lexical diversity. The SISAT results coincide with the findings of the linguistic analysis, validating the correspondence between institutional performance and writing patterns.

3.3. General Trends in Narrative Corpus

The highest lexical density was observed in the second period, suggesting texts with greater semantic load. The third period shows greater thematic and expressive diversity, although with slightly lower lexical density, indicating a transition towards more descriptive and experiential narratives. The corporation reflects a coherent youth narrative universe, in which writing fulfills affective, identity, and learning functions.

3.4. Characteristics of the Private University Corpus

The corpus included 74 narratives, distributed across three academic periods. Lexical density ranged from 22% to 39.6%, and vocabulary richness varied according to the period, reflecting an evolution towards more diverse and expressive narratives (Table 4).

Table 4. Relationship between lexical metrics and writing performance (SISAT) by academic terms.

Period	Number of Texts	Tokens/Words	Unique Words	Lexical Density (%)	Average SISAT Score (%)
First	24	12,483	4952	39.6	13.6
Second	25	19,842	4378	22.1	13.7
Third	25	22,846	6192	27.1	14.2

The higher lexical density in the first period reflects concentrated and symbolic vocabulary; the third period combines lexical richness with more elaborate narrative structures. The average SISAT score shows stability with a slight increase in the third period.

3.5. Linguistic and Thematic Patterns

Nouns related to childhood, family, time, and values predominate, evidencing an affective, moral, and formative approach. Recurring thematic nuclei include familial–affective, symbolic–existential, social–everyday, and moral–formative. The narrative has a tripartite structure (beginning → conflict → lesson), with a predominance of the imperfect and perfect past tenses, simple temporal connectors and reflective closure at intermediate levels. The most frequent rhetorical figures were metaphor, personification, anaphora, hyperbole and antithesis, contributing to a style of symbolic and moralizing realism (Table 5).

Table 5. Relationship between SISAT levels and dominant linguistic features.

SISAT Level	Linguistic Features	Quantitative Evidence
Requires support	Weak cohesion, limited lexicon, fragmented narration	Low lexical density (≈20–22%), simple sentences, abrupt endings
Developing	Partial cohesion, concrete lexicon, basic connectors	Medium lexical density (≈22–36%), recurrent thematic sentences, brief reflective endings
Expected level	Clear cohesion, varied and symbolic lexicon, functional connectors	High lexical density (≈30–40%), thematic diversity, reflective endings, use of rhetorical figures

Most students are at the “Developing” level, with functional narrative skills but room for improvement in cohesion, lexical variety and discourse complexity. The correlation between SISAT and linguistic analysis confirms that the instrument reflects performance patterns consistent with textual quality.

3.6. General Trends in Narrative Corpus

The public university has a higher text volume, while lexical density is higher in the private university during the first period. The average SISAT score shows slightly higher trends in the private institution (Table 6 and Table 7).

Table 6. Comparison of lexical metrics and SISAT performance by institution.

Period	Institution	N texts	Tokens/Words	Unique Words	Lexical Density (%)	Average SISAT Score
First	Public	24	27,486	5912	21.5	13.58
Second	Public	25	19,483	5762	29.6	13.7
Third	Public	25	21,584	4872	22.0	12.5
First	Private	24	12,483	4952	39.6	13.6
Second	Private	25	19,842	4378	22.1	13.7
Third	Private	25	22,846	6192	27.1	14.2

Table 7. Spearman correlations between linguistic indicators and SISAT scores by type of university.

Institution	Indicator	ρ (Spearman)	Interpretative Trend
Public	Unique words	+0.50	Moderate positive: higher lexical richness associated with SISAT score
Public	Lexical density	+0.50	Moderate positive: lexical density associated with score
Public	Total tokens	−0.50	Moderate negative: higher volume does not imply better performance
Private	Unique words	+1.00	Perfect positive: lexical diversity clearly related to SISAT
Private	Total tokens	+1.00	Perfect positive: higher text production associated with score
Private	Lexical density	−0.50	Moderate negative: excessive lexical density reduces score

The justification for calculating correlations separately is that, as indicated in the Methods section, SISAT scores are ordinal, some linguistic variables are not normally distributed, and the sample size per period is limited. Therefore, Spearman’s correlation was used. Correlations were calculated separately for public and private universities to explore whether the relationship between linguistic characteristics and performance varied according to institutional context, given that student profiles and educational practices differ between types of institutions. In the private university, some linguistic variables (unique words and text volume) varied linearly with SISAT scores in this sample, resulting in a perfect correlation of ρ = 1.00. Although uncommon, this outcome reflects the homogeneity of narrative development in this institution for the analyzed period, considering the limited sample size (n = 74). This strategy allows the results to be interpreted in a contextualized manner, rather than assuming a homogeneous relationship for both groups (Table 8).

Table 8. Kruskal–Wallis comparisons of SISAT scores across periods and institutions.

Comparison	H	gl	p	Interpretation
Periods (both institutions)	0.17	2	>0.05	No significant differences; slight descriptive trend between periods
Institutions (average score)	1.33	1	>0.05	No significant differences; descriptively, private slightly higher

At the aggregate level, no statistically significant differences are observed between periods or institutions. However, descriptive and correlational analyses show distinct institutional patterns in the relationship between lexical richness, lexical density, textual volume, and SISAT performance. This highlights the importance of combining statistical and descriptive analyses to interpret the development of creative writing according to institutional context.

4. Discussion

The analysis of narrative corpora from students at a public and a private university allows us to reflect on academic writing in higher education, integrating rhetorical, linguistic, sociocultural, cognitive, and didactic perspectives. The use of corpus linguistics as a systematic and empirical framework facilitates the identification of lexical–grammatical, discursive, and rhetorical patterns associated with different levels of performance (Biber et al., 1999; Granger, 2024; Hyland, 2016, 2019; McEnery & Hardie, 2012). Metrics such as lexical density and richness, text volume, syntactic complexity, and thematic progression provide quantitative evidence of writing proficiency and its relationship with institutional criteria (Barbara et al., 2024; Biber, 2012; Lusta et al., 2023; Ten Peze et al., 2024; Ueno & Takeuchi, 2023).

The use of ad hoc corpora and automated techniques based on artificial intelligence enables efficient tokenization, tagging, and calculation of linguistic indicators, integrating institutional performance analysis with pedagogical evidence (O’Donnell et al., 2025). This demonstrates how automated assessment can complement teaching by providing objective and timely feedback without replacing pedagogical judgment.

From the perspective of cognitive psychology, writing involves planning, textualization, and revision mediated by working memory and prior knowledge (Flower & Hayes, 1981). The results show that lexical richness and density are associated with performance on the SISAT, although patterns vary according to institutional context: in public universities, a greater volume of text does not guarantee high scores, while in private universities, textual diversity and length are related to better results, reflecting differences in self-regulation strategies (Zimmerman, 2000).

From a rhetorical perspective, thematic progression, strategic organization, and the use of discourse markers reveal the audience’s expectations and the author’s position (Hyland, 2002; Swales, 1990). Automated assessment that identifies these patterns allows for a comprehensive evaluation of writing. Lexical richness and the use of connectors, consistent with systemic functional linguistics and corpus studies (Biber et al., 1999; Halliday, 1985), stand out as key indicators. Texts from public universities show greater volume but lower lexical density, while those from private universities tend to be more cohesive and denser, influenced by the sociocultural context (Barton & Hamilton, 1998; Street, 1984).

Narrative and creative writing strengthen self-regulation, symbolic expression, and ethical reflection (Barbara et al., 2024; Bruner, 2001). The application of automated techniques offers immediate and objective feedback, although it must be complemented by teacher assessment to preserve creativity and stylistic diversity.

Taken together, these findings show that academic writing develops gradually and is mediated by multiple factors. Although the study is limited to two universities and does not allow for broad generalizations, it provides novel evidence on how institutional context influences the relationship between textual characteristics and assessment performance. The results suggest the need for differentiated pedagogical strategies and the integration of automated assessment with qualitative analysis, promoting expressive, critical, and reflective skills in various university contexts.

5. Conclusions

The differences in the relationship between text length, lexical richness, and writing performance between the two universities suggest that the institutional context shapes how writing skills develop in university students, providing evidence that academic writing depends not only on individual skills but also on educational and sociocultural factors. Lexical diversity and textual cohesion emerge as consistent indicators of performance, validating the usefulness of automated and corpus metrics to complement institutional assessment by providing objective data on areas for improvement in academic writing. Narratives show that writing fulfills expressive, reflective, and formative functions beyond institutional assessment, suggesting that narrative work can strengthen self-regulation, symbolic expression, and ethical reflection by integrating the academic dimension with the personal and social dimensions of learning. Although this study is limited to two universities, its findings offer insights applicable to the teaching of university writing, particularly regarding how to adapt pedagogical strategies to different institutional contexts and how to integrate automated tools with qualitative analysis to strengthen writing competence.

Appendix A.1. SISAT Scale for Written Text Exploration (SEP, 2019)

The evaluation of student narratives was conducted using the SISAT scale, designed to explore the quality of written texts in higher education. Performance is classified into three levels based on total score: Expected Level—15–18 points; Developing Level—10–14 points and Requires Support—9 points or less. The scale evaluates six dimensions: readability, fulfillment of communicative purpose, relationship between words and sentences, lexical diversity, correct use of punctuation, and proper application of spelling rules. Internal consistency is α ≈ 0.74, indicating reliability.

Table A1. SISAT Scale for written text exploration.

Dimension	Score 3	Score 2	Score 1
I. Readability	Legible: correct word separation, hyphenation, and letter spelling	Moderately legible: substitution, omission, or addition errors; partial word separation	Illegible: pre-alphabetic writing; incorrect separation; unreadable
II. Communicative Purpose	Clear, organized, and coherent message	Partially clear: incomplete or mixed ideas; missing components	Not clear: message confused, sequence lost, no proper organization
III. Word–Sentence Relationship	Correct verb tenses, gender, number; variety of connectors	Some errors in tenses, gender, number; limited connectors	Incorrect use of tenses, gender, number; no connectors
IV. Lexical Diversity	Rich, varied, and context-appropriate vocabulary	Limited or repetitive; some inappropriate words	Very reduced or irrelevant vocabulary; minimal text production
V. Punctuation	Correct use of period, comma, and other required marks (?, !, quotes)	Partial use, some omissions	Punctuation absent or incorrect throughout
VI. Spelling Rules	Correct capitalization, accents, and letters representing sounds	Minor errors; incorrect accentuation in common words	Serious errors: missing capitals, accents, incorrect letters

Note: Total score is the sum of all dimensions (3, 2, or 1 point per dimension).

Table A1. SISAT Scale for written text exploration.

Dimension	Score 3	Score 2	Score 1
I. Readability	Legible: correct word separation, hyphenation, and letter spelling	Moderately legible: substitution, omission, or addition errors; partial word separation	Illegible: pre-alphabetic writing; incorrect separation; unreadable
II. Communicative Purpose	Clear, organized, and coherent message	Partially clear: incomplete or mixed ideas; missing components	Not clear: message confused, sequence lost, no proper organization
III. Word–Sentence Relationship	Correct verb tenses, gender, number; variety of connectors	Some errors in tenses, gender, number; limited connectors	Incorrect use of tenses, gender, number; no connectors
IV. Lexical Diversity	Rich, varied, and context-appropriate vocabulary	Limited or repetitive; some inappropriate words	Very reduced or irrelevant vocabulary; minimal text production
V. Punctuation	Correct use of period, comma, and other required marks (?, !, quotes)	Partial use, some omissions	Punctuation absent or incorrect throughout
VI. Spelling Rules	Correct capitalization, accents, and letters representing sounds	Minor errors; incorrect accentuation in common words	Serious errors: missing capitals, accents, incorrect letters

Note: Total score is the sum of all dimensions (3, 2, or 1 point per dimension).

Appendix A.2. Narrative Examples and Linguistic Analysis

Appendix A.2.1. Example 1: Developing Level (Public University)

Excerpt 1 (SISAT: Developing Level): “Es una niña llamada Isabella, procedente de una familia con bajos ingresos, que quería tener un gran futuro para ayudar a sus padres. La pequeña Isabella tenía muchos sueños que quería cumplir. Un día decidió esforzarse para cambiar su vida y apoyar a su familia, porque sabía que su futuro dependía de ello.”

Linguistic and Discursive Analysis: Clear narrative purpose and thematic continuity focused on family and personal aspiration. Recurrent orthographic errors and limited punctuation.

• Extended syntactic structures with weak segmentation.
• Moderate lexical density, with repetition of general verbs (quería and tener).
• Cohesion relies mainly on chronological sequencing, without explicit connectors.

SISAT Level: Developing.

Appendix A.2.2. Example 2: Expected Level (Private University)

Excerpt 2 (SISAT: Expected Level): “Desde la infancia, Troy sintió que el mundo se extendía más allá de las montañas que rodeaban su pueblo. Mientras ayudaba a su abuelo en el campo, imaginaba caminos que lo conducirían a otros lugares. Al regresar a casa, comprendió que la verdadera magia no estaba en los objetos, sino en la capacidad de imaginar y decidir su propio destino.”

Linguistic and Discursive Analysis:

• Clear cohesion, varied and precise vocabulary, controlled syntactic structures.
• Medium–high lexical density, supported by abstract nouns and specific verbs (comprendió, imaginaba, and conducirían).
• Temporal and cohesive connectors enhance thematic progression.
• Reflective closure integrated naturally into the narrative.

SISAT Level: Expected.

Appendix A.3. Comparative Summary of Linguistic Features

Feature	Developing Level	Expected Level
Narrative Purpose	Clear but basic	Clear and rhetorically sustained
Lexical Density	Moderate	Medium–high
Lexical Richness	Limited, repetitive	Varied and precise
Cohesion	Chronological	Explicit cohesive devices
Syntax	Long, weakly segmented clauses	Balanced coordination and subordination
Orthography	Frequent inaccuracies	Mostly correct
Reflective Closure	Explicit and simple	Integrated and elaborated

Appendix A.4. Pedagogical Relevance

These examples illustrate how narrative writing quality varies according to performance level and institutional context. The contrast between Developing and Expected levels supports the study’s quantitative findings, showing that automated linguistic indicators such as lexical density and richness meaningfully correspond to institutional assessment. Including narrative exemplars enhances transparency and provides concrete pedagogical insight into the developmental nature of academic writing in higher education.

Appendix C.1. Automated Linguistic Analysis

Automated linguistic analysis was conducted using text processing techniques, including tokenization and lexical normalization. These procedures enabled the calculation of quantitative linguistic indicators such as lexical density, lexical richness, and the frequency of words and grammatical categories, as well as the identification of frequent words, key phrases, and thematic categories.

The analysis was carried out using custom-developed online tools to ensure consistency, transparency, and replicability of the evaluation process.

Appendix C.2. Online Tools

Appendix C.2.1. Linguistic Pattern Evaluator

The first tool, Linguistic Patterns, is an online linguistic pattern evaluator developed by one of the researchers and is fully available in Spanish. This tool performs lexical and grammatical analysis of written texts and provides automated metrics related to lexical density, lexical richness, and thematic distribution.

URL: https://chatgpt.com/g/g-682b7d9107d88191850ccd0c8e8c58a7 (accessed on 23 January 2026).

Figure A1. Linguistic pattern evaluator interface.

Figure A1. Linguistic pattern evaluator interface.

Appendix C.2.2. SISAT Evaluator

The second tool is the SISAT evaluator, also developed by one of the researchers and available entirely in Spanish. This system processes texts extracted from PNG images.

As shown in Figure A3, the system first determines whether the input text is handwritten. If handwriting is detected, the system evaluates text legibility. Subsequently, the text is assessed using the corresponding SISAT rubric, and an automated evaluation report is generated.

URL: https://chatgpt.com/g/g-67e1b0bffcb4819198589c95fb6f7ce7 (accessed on 5 July 2025).

Figure A2. SISAT evaluator interface.

Figure A2. SISAT evaluator interface.

Figure A3. Flowchart of the SISAT evaluation model.

Figure A3. Flowchart of the SISAT evaluation model.

Appendix C.3. Study Variables

The study variables were classified into institutional and linguistic variables.

• Institutional variable:

  ○

  SISAT performance level (Requires support, In development, and Expected level).

• Linguistic variables:

  ○

  Lexical density.

  ○

  Lexical richness.

  ○

  Frequency of grammatical categories.

  ○

  Number of sentences.

  ○

  Thematic complexity.