Introduction

Translation and localization are connected through their shared goal of facilitating cross-cultural communication, with localization evolving as a specialized form of translation adapted to technological and commercial contexts. The professionalization of localization has influenced translator training programs and industry practices, making it an integral part of the modern translation ecosystem (Folaron, 2008). However, they are separated by theoretical approaches, as translation studies have traditionally focused on broader cultural and linguistic dimensions and considered these inseparable (O’Hagan and Mangiron, 2013), whereas the localization process has been examined separately in its linguistic, technical, and cultural dimensions (Fry, 2003). The study of video game localization further complicates this relationship by introducing new dimensions of cultural adaptation and technical complexity that challenge existing theoretical frameworks (Mangiron, 2006).

The simplification hypothesis has primarily been tested as a translation universal in translation studies (Baker, 1993). Researchers have focused on distinguishing translated text from non-translated text (Laviosa, 1998; Ferraresi et al., 2018; Liu et al., 2022), interpreted text (Bernardini et al., 2016), EFL writing (Chen et al., 2024), and native language texts (Kruger & Van Rooy, 2016; Wang et al., 2024) through lexical or syntactic indices borrowed from computational and quantitative linguistics. However, owing to the similarity between localization and translation, to the best of our knowledge, there has been no quantitative comparison of localized and native texts to investigate the simplification hypothesis in video game localization.

To address the gap, we adopt Baker’s (1993) definition of simplification to test the linguistic complexity of localized game texts. The remainder of this article proceeds as follows. Section II reviews (a) the relationship between translation and localization and (b) the literature on simplification in translation studies. Section III presents the method of using entropy and mean dependency distance to measure linguistic complexity, describes the compilation of the corpus, and explains why Black Myth: Wukong (China), Sekiro: Shadows Die Twice (Japan), and Red Dead Redemption 2 (USA) were selected for comparison. Sections IV and V provide the descriptive results and a prescriptive discussion. Section VI presents some concluding remarks.

Literature review

Translation versus localization

The ambiguous relationship between the translation studies concepts of “localization” and “translation” is largely due to a lack of theorization of localization (O’Hagan and Mangiron, 2013). The two terms are often used interchangeably but they represent distinct processes and each has distinct implications for global communication and digital content adaptation. Translation is traditionally defined as the process of converting text from one (source) language to another (target) language while preserving its meaning and intent. It is a linguistic activity with a long history dating back to ancient civilizations. Newmark (1998) described translation as “the craft or art of rewriting, correctly, completely, and idiomatically, a text from one language into another.” Translation has been applied to literature, religious texts, and scientific works and plays a crucial role in cross-cultural communication and knowledge transfer.

Localization is a more recent concept, having emerged with the globalization of software and digital content. It involves adapting a product or service to meet the linguistic, cultural, and technical requirements of a specific target market. The Localization Industry Standards Association defines localization as “the process of adapting a product to account for differences in distinct markets” (Folaron, 2008). This process goes beyond mere translation to encompass adjustments to user interfaces, date and number formats, currency symbols, and other culturally specific elements. Localization is particularly critical in the software and video game industries, in which products must be tailored to different regions to ensure usability and market acceptance.

The historical development of translation and localization reflects their evolving roles in global communication. Translation has been practiced for centuries: a notable historical example is the preparation of the Greek Septuagint by translation from the Hebrew during the Hellenistic period (Schäler, 2010). The development of translation theory has been marked by the adoption of various approaches, including the distinction between literal and free translation and the debate between domestication and foreignization (Venuti, 1995). Translation studies emerged as an academic discipline in the 20th century, with Bassnett (2007) and Lefevere (2002) contributing to its theoretical framework. Localization emerged as a distinct practice in the late 20th century with the rise of the software industry. The term was coined by software developers in the late 1980s to describe the process of adapting software products for international markets (Folaron, 2008). Initially, localization efforts were focused on addressing language-specific requirements, such as character encoding and text directionality. As the industry evolved, the scope of localization was expanded to include cultural and technical adaptations, such as modifying user interfaces and supporting local file formats. The growth of the internet and digital media further broadened the scope of localization to encompass websites, multimedia content, and online services.

There are key differences between translation and localization in their scope, focus, and process. Whereas translation is primarily concerned with the linguistic conversion of text, focusing on preserving the meaning and intent of the source content, localization addresses a broader range of cultural, technical, and functional issues. Translation is therefore a component of the more comprehensive adaptation process of localization. In the video game context, for example, localization may include modifying game mechanics, adjusting difficulty levels, and altering visual and auditory elements to suit the target audience (O’Hagan & Mangiron, 2013). Accordingly, whereas the process of translation typically involves a linear workflow of converting text from the source language to the target language, often with minimal contextual or technical constraints, localization involves a more complex and iterative workflow that often includes internationalization, translation, cultural adaptation, and quality assurance testing (Esselink, 2000). The localization process also requires close collaboration between multiple stakeholders, including developers, designers, and localizers, to ensure that a product is adapted effectively for the target market. Localization can be further distinguished from translation by its strong emphasis on technical aspects, such as software engineering and user interface design, and its use of specialized tools and technologies, such as computer-assisted translation tools, translation memory systems, and localization platforms. These tools facilitate the extraction and translation of text and its reintegration into the software product. Although translation has also benefited from technological advancements, it does not typically involve as high a level of technical complexity as localization (Gouadec, 2007).

The video game industry provides a rich context for examining the relationship between translation and localization. Early video games, such as Pong in 1972 and Space Inaders in 1978, had minimal text and required little localization. As games became more complex and narrative-driven, the need for localization grew. The 1980s saw the emergence of Japanese arcade games and home consoles, which were often localized for Western markets. For example, Nintendo’s Famicom was launched in Japan in 1983 and then released as the Nintendo Entertainment System in the United States in 1985 (O’Hagan & Mangiron, 2013). However, the interactive nature of video games presents unique challenges for localization. Games often contain large amounts of text, including dialogue, instructions, and narrative elements, that must be translated and adapted for the target audience. Furthermore, games may have culturally specific content, such as historical references, religious symbols, and representations of social norms, which require careful consideration during localization. Further challenges can be posed by technical constraints, such as limited storage capacity and processing power, which were particularly severe in the early period of gaming (Chandler, 2005).

Game developers and localizers have adopted various strategies to address these challenges. One common strategy is internationalization, which involves designing games with localization in mind from the outset and can include using standardized character sets, separating text from code, and supporting multiple languages. Another strategy is cultural adaptation, which involves modifying game content to suit the cultural preferences and sensitivities of the target audience and can include adjusting character designs, altering dialogue, and modifying gameplay mechanics (O’Hagan & Mangiron, 2013). The importance of technology in localization cannot be overstated. The localization process has been revolutionized by computer-assisted translation tools, which automate many of the repetitive tasks involved in translation through such features as translation memory, which stores previously translated segments for reuse, and terminology management, which ensures consistency in the use of specialized terms. These tools also facilitate collaboration between translators and localizers, allowing them to work on different parts of a project simultaneously (Gouadec, 2007). The advent of machine translation (MT) and artificial intelligence (AI) has further transformed localization practices. MT systems, such as Google Translate, can quickly translate large amounts of text, although the output often requires post-editing by human translators to achieve an acceptable level of quality.

Simplification in translation studies

One of the most widely debated and investigated hypotheses in translation studies is the simplification hypothesis, which proposes that translators instinctively streamline the linguistic form and content when rendering texts from one language to another. This hypothesis suggests that translated language constitutes a distinct variety of the target language and is marked by specific patterns of lexical, syntactic, and stylistic simplification that differentiate it from non-translated, or “native,” texts (Baker, 1993; Laviosa, 1998a). At its core, simplification can be understood as a subconscious process through which translators reduce complexity in response to cognitive constraints and communicative demands, resulting in texts that are arguably easier for target language readers to process than the source language text is for source language readers to process (Blum-Kulka & Levenston, 1978).

Early empirical investigations into simplification were focused primarily on the lexical dimension and identified several characteristic features of translated texts, including a lower proportion of content words than of grammatical words; a higher frequency of common vocabulary items than of rare or specialized terms; and a reduced type-token ratio, indicating less lexical diversity (Laviosa, 1998; Olohan, 2004). It was also observed that translators frequently replace formal or archaic source text vocabulary with more colloquial equivalents, effectively “modernizing” the lexical profile of the text (Blum-Kulka & Levenston, 1978). These patterns were initially documented through manual analyses of relatively small text samples, with studies often limited to specific language pairs or text types, raising questions about the generalizability of the findings.

The extension of simplification research to syntactic features revealed additional distinctive patterns in translated texts. It was found that translators systematically transform complex syntactic structures into simpler ones, often replacing finite clauses with non-finite constructions and favoring shorter sentences (Malmkjær, 1997; Vanderauwera, 2022). These syntactic adaptations were interpreted as strategies to reduce processing demands on readers, aligning with broader principles of communicative efficiency in language use. The methodological landscape of simplification research underwent a significant transformation with the advent of corpus-based approaches, which allowed for more systematic comparisons of translated and non-translated texts. Baker’s (1993) pioneering work established the comparable corpus method, enabling researchers to identify patterns across large text collections and apply statistical techniques to quantify differences between text types.

Despite substantial evidence supporting the simplification hypothesis, the findings of many studies have challenged its universal applicability. Xiao and Yue (2009) documented greater average sentence lengths in translated texts compared with their corresponding source texts, and Ferraresi et al. (2018) found that translated texts exhibited greater lexical complexity than their non-translated counterparts. These contradictory results have prompted scholars to reconsider what factors might modulate simplification patterns, with a growing recognition of the significant influence of such variables as the genetic proximity of the source and target languages, the text genre, and translation directionality.

Methodological advancements have addressed some limitations in the early research through the adoption of more sophisticated analytical techniques. The integration of multivariate statistical methods has enabled researchers to account for multiple linguistic features simultaneously, providing a more nuanced understanding of how simplification interacts with other translation phenomena (Liu et al., 2022). Another promising development is the application of entropy as a quantitative measure of linguistic complexity, as this offers a rigorous framework for analyzing simplification across different text types and language pairs (Liu et al., 2022).

The evolution of research on the simplification hypothesis reflects broader trends in translation studies toward greater methodological rigor and theoretical sophistication. Whereas the early investigations relied on intuitive observations and small-scale analyses, more recent studies have increasingly adopted computational tools and large-scale corpora to explore simplification across diverse linguistic contexts. This has not only strengthened the empirical foundations of the hypothesis but also revealed the complexity of simplification as a dynamic phenomenon shaped by multiple interacting factors.

Looking forward in this fundamental area of translation studies, several avenues for research present themselves. There remains a lack of systematic investigations into how simplification varies across different text genres, particularly in the wake of evidence that creative or informal texts may exhibit different simplification patterns than informational or technical texts (Kruger & Van Rooy, 2016). There is also a need for cross-linguistic studies involving genetically distant language pairs to determine whether the cognitive pressures underlying simplification operate differently in translation between languages with distinct structural properties. Finally, the potential for digital humanities approaches to uncover new dimensions of simplification through the analysis of massive text collections offers exciting possibilities for future discoveries.

Calculating linguistic complexity through entropy- and dependency-based measures

Entropy, originating from Shannon’s information theory (1948), serves as a crucial metric for uncertainty and information content within datasets. In linguistics, it measures both the frequency and distribution of linguistic units, thus offering insights into various aspects of language complexity. The calculation of entropy involves assessing the probabilities of different outcomes or messages in a dataset, with higher entropy indicating greater uncertainty or information content. Entropy has been used as a metric to effectively quantify the complexity of morphological systems by considering the number of states (morphological forms) and their frequency distribution (Baerman, 2012; Ackerman & Malouf, 2013). At the lexical level, entropy measures the richness and variety of vocabulary in texts by analyzing word frequency distributions, with higher entropy values indicating more diverse and sophisticated lexical choices (Juola, 1998, 2008, 2013). The application of entropy in linguistic research has evolved through several key studies. Genzel and Charniak (2002) introduced the “constancy rate principle,” showing how local entropy increases with sentence number, and Tanaka-Ishii (2005) explored how token uncertainty can help in determining context boundaries. Subsequent studies building on these foundations demonstrate the applicability of entropy across various linguistic domains (Mehri & Darooneh, 2011; Suo et al., 2012; Yang et al., 2013; Bentz et al., 2017; Lowder et al., 2018; Friedrich et al., 2020; Friedrich, 2021). In translation and interpreting research, entropy has gained significant attention as a tool for understanding cognitive and linguistic processes. Wei (2022) used surprisal and entropy as cognitive load metrics and found the former effective for predicting translation production time and the latter effective for predicting source text reading time. Chen, Liu, and Altmann (2017) used an entropy analysis to demonstrate the unique linguistic profiles of different text types, and Yerkebulan et al. (2021) developed an entropy-based approach to detect patterns in multilingual texts. Liu et al. (2022) showcased the effectiveness of entropy-based methods in examining translation universals in Chinese texts.

Dependency distance (DD), a term coined by Heringer et al. (1980) and then extended by Hudson (1995), represents the number of words intervening between two syntactically related words, or the difference in their linear positions. This measure can quantify syntactic complexity by assessing the distance between related words in a sentence, reflecting the cognitive load during language processing. A smaller DD typically indicates less cognitive demand, as it requires less working memory capacity to maintain the relationship between words. The calculation of DD involves analyzing the syntactic relationships in a text through dependency parsing to identify connections between words and measure the distance between them. The average DD across a text can then be computed, with lower values suggesting simpler syntactic structures from a processing perspective.

Minimizing DD is a universal characteristic of natural languages. Liu (2008) calculated the average DD of 20 languages and discovered a DD threshold for most human languages of no more than three words to satisfy working memory constraints. This finding was supported by Futrell et al. (2015), who analyzed a corpus of 37 diverse languages and found that the overall DD for all these languages was shorter than the random baseline, thus confirming DD minimization as a cross-linguistic universal. Lei and Wen (2020) observed a tendency toward DD minimization in the diachronic variation of State of the Union addresses over two hundred years.

In translation studies, DD has proven valuable for investigating distinctive syntactic patterns. Fan and Jiang (2019) found that translated English texts have a longer mean dependency distance (MDD) than native English texts, indicating greater cognitive demands in translation. Similarly, Liang and Sang (2022) observed a longer DD in English abstracts translated from Chinese than in abstracts originally written in English, and attributed this finding to translation cognitive costs and source language influence, particularly noting that Chinese has a longer DD than English, which results in the syntactic characteristics of the source language “shine through” in translated English texts (Teich, 2003).

Interpreting research presents interesting variations. Liang et al. (2017) found different patterns in three types of interpreting: consecutive interpreting, simultaneous interpreting, and read-out translated speech. Although the shorter DD in interpreted speeches potentially indicates higher cognitive loads due to memory constraints, the results showed that read-out translated speech had the longest DD and consecutive interpreting the shortest. This suggests that interpreters may strategically reduce DD to manage cognitive load, aligning with the universal language property of DD minimization. Yan and Liang (2022) found that the DD of students’ consecutive interpreting was shorter when they were experiencing greater foreign language anxiety, indicating a higher cognitive load during interpreting tasks.

The cognitive interpretation of DD relates to how our brains process language. Hudson (1995) and Liu et al. (2017) noted that DD reflects the dynamic cognitive load during language processing, with longer distances potentially increasing processing difficulty. This aligns with dependency locality theory (Gibson, 1998, 2000), which suggests that processing difficulty increases with the number of unresolved dependencies. Similar concepts for measuring processing difficulty, such as the principle of early immediate constituents, have been used by some phrase structure grammars (Hawkins, 1994, 2004).

Based on the above review, the research questions are as follows:

  1. 1.

    How does the linguistic complexity of localized video game texts compare with that of original texts in terms of lexical and syntactic features?

  2. 2.

    Does the localization of video game texts involve the simplification of linguistic features?

  3. 3.

    What are the implications of the findings for the broader understanding of translation universals and the practice of localization?

Method and materials

Word entropy as an index of lexical complexity

To measure linguistic complexity, we employed Shannon entropy and Mean Dependency Distance (MDD) as our primary indices. These measures were chosen over traditional metrics such as the type-token ratio or average sentence length for their greater sensitivity to subtle variations in syntactic structure. The type-token ratio, while useful for assessing lexical diversity, does not capture the intricacies of syntactic complexity. Similarly, average sentence length provides a general indication of complexity but fails to account for the internal structure of sentences. In contrast, Shannon entropy quantifies the unpredictability of linguistic structures, offering a more nuanced measure of complexity. MDD, on the other hand, evaluates the average distance between syntactically related words, providing a direct measure of syntactic complexity. These indices are particularly well-suited for analyzing the nuanced syntactic changes that occur during the localization process, allowing for a more precise assessment of simplification phenomena.

Word entropy was used in this study to measure lexical complexity. Entropy, originating from information theory, was developed by Shannon (1948, 1951) as a measure of randomness and uncertainty. Shannon’s entropy H can be computed as follows:

$$H(X)=-\mathop{\sum }\limits_{i=1}^{n}Pi\log \,bPi$$
(1)

in which H(X) is the information entropy of a random variable (X) and represents its average uncertainty; Pi is the probability of a certain word type in the text, for calculating its relative frequency; and logn Pi takes the logarithm of the probability Pi based on b (when b equals 2, the unit of information is bits), reflecting the amount of information provided by the value i. The entropy of a text is calculated as the sum of expected values for all types.

For example, in the sentence “The monkey leveled the Court,” “the” appears twice, so it is shown as \(\frac{2}{5}{\log }_{2}\frac{2}{5}\) and the remaining words only once as \(\frac{1}{5}{\log }_{2}\frac{1}{5}\); therefore, the entropy of this sentence would be computed as in Fig. 1.

Fig. 1
figure 1

Formula for calculating word entropy of the sample sentence “The monkey leveled the Court.”.

Full size image

Mean dependency distance as an index of syntactic complexity

We used DD to measure syntactic complexity. This method is grounded in the dependency relations between individual words, as discussed by Tesnière (1959), Hudson (2007), and Liu (2009). A dependency relation is widely recognized to have three core properties, as outlined by Tesnière (1959) and Liu (2009): it is (i) a binary relation between two linguistic units; (ii) always asymmetrical and directed, with one unit serving as the head and the other as the dependent; and (iii) labeled, with the label indicating the type of the dependency relation typically placed above the arc that connects the two units. By utilizing these three properties, a syntactic dependency tree or a directed dependency graph can be constructed to represent the syntactic structure of a sentence. The directed acyclic graph in Fig. 2 represents the dependency structure of the sentence “The monkey leveled the Court.” Each word is connected through dependency relations, with labeled arcs extending from governors to dependents (Liu, 2008). The direction of these arcs demonstrates the asymmetrical nature of the relationship between the two connected units. The number beneath each word indicates its position within the sentence, which is essential for calculating DD. Liu et al. (2009) introduced a method for measuring DD in sentences and texts in which DD is determined by the difference in positions between connected words. DD is negative when the governor precedes the dependent and positive when the governor follows the dependent; however, the absolute value of DD is typically used for measurement purposes.

Fig. 2: Graph showing the dependency relations of a sample sentence.
figure 2

In a dependency relationship between two words, one is referred to as the dependent (DET) and the other as the governor. An arc labeled with the dependency type extends from the governor to the dependent, as described by Liu in 2008. This directed arc highlights the asymmetrical nature of the relationship between the governor and the dependent. Additionally, the numbers beneath the words denote their linear positions within the sentence, which are crucial for calculating the distance between dependents and governors (DD).

Full size image

The MDD of a text can be computed as:

$$MDD=\frac{1}{n-s}\mathop{\sum }\limits_{i=1}^{n-s}|DDi|$$
(2)

in which n is the number of words and s the number of sentences in the sample. DDi is the DD of the i-th syntactic link of the text. For the example sentence “The monkey leveled the Court,” by subtracting the position number of a word from that of its corresponding governor, the following sequence of DDs can be derived: 1, 1, 0, 1, 2. Following Formula (2), the MDD for this sentence is 5/4, resulting in a value of 1.25.

Corpora

In line with our goal of investigating the simplification hypothesis by comparing the linguistic complexity of L2 (localized) with L1 (original) video games, we selected three popular video games—Black Myth: Wukong (L2), produced in China; Sekiro: Shadows Die Twice (L2), produced in Japan; and Red Dead Redemption 2 (L1), produced in the USA—for a comparative analysis of linguistic complexity. Several key criteria were used to confirm the comparability of these three titles. All are rated as Triple-A (3 A) productions, which are characterized by large development budgets, high production values, and significant marketing effort. Furthermore, each falls within the action genre and features rich narratives and expansive game-worlds: Black Myth: Wukong draws on Chinese mythology to tell a story filled with fantasy and adventure; Sekiro: Shadows Die Twice is set against the backdrop of Japan’s Warring States period, blending historical and mythological elements; and Red Dead Redemption 2 immerses players in an open world of the American West, allowing them to experience the lives and adventures of cowboys.

Three sub-corpora—Black_myth, Sekiro, and RDR2—were compiled by extracting the voice-over and dialogue in computer-generated animation (see Table 1). As Black Myth: Wukong had the smallest text size, we used it as the benchmark and controlled the other two games to match its size. The text files were randomly shuffled, cleaned, and manually checked before making the calculations.

Table 1 Summary of corpora.
Full size table

Results

A visualization was generated for an initial exploration of the linguistic complexity universals in popular video games. The scatter plot in Fig. 3 displays the entropy and MDD values for the three game sub-corpora. Each game cluster is centered at mean values marked by black crosses, with dashed rectangles indicating data boundaries. Distinct clustering is evident for each game but there is notable overlap in the Entropy–MDD space. The overlap is more pronounced between Black Myth and Sekiro and less pronounced between Red Dead Redemption 2 and the others, revealing shared patterns and unique characteristics in linguistic complexity across the three games. Shapiro–Wilk and Levene tests were conducted to assess the normality of the data before proceeding with inferential analysis. Tukey’s post-hoc tests were conducted to investigate the differences among the two measures for the three groups, with an alpha level set at 0.05 for all statistical tests.

Fig. 3
figure 3

Scatter plot with group centers and boundaries.

Full size image

Comparing the entropy of the three sub-corpora

The entropy values for the texts in the three sub-corpora are visualized in Fig. 4, and a detailed statistical summary of these results is shown in Table 2. The MDD values for the three sub-corpora were similar, with minor differences: Black_Myth had the highest mean value (M = 6.354, SD = 0.124), followed by Sekiro (M = 6.353, SD = 0.133) and then RDR2 (M = 6.032, SD = 0.123). Regarding entropy, a one-way analysis of variance (ANOVA) was performed to determine whether the patterns of variation in entropy across the three sub-corpora were similar from a statistical standpoint. Given that each group had more than 30 dependent variables (N = 219), the assumptions of a normal distribution (Shapiro–Wilk test) and homogeneity of variances (Levene’s test) were validated (p > 0.05), as shown in Table 2. The ANOVA test result (F = 471.219, p < 0.05) indicated a significant difference between the mean entropy values of the three sub-corpora. Therefore, we rejected the null hypothesis that the means are equal and proceeded to conduct Tukey’s post-hoc tests to identify the differences. The findings of our Tukey’s post-hoc tests are summarized in Table 3 and indicate that the entropy of RDR2 was significantly different from those of Black_Myth (p < 0.05) and Sekiro (p < 0.05) but there was no significant difference in entropy between Black_Myth and Sekiro. Furthermore, the differences in entropy value between the L1 game sub-corpus (RDR2) and each L2 game sub-corpus (Black_Myth and Sekiro) were larger than the difference between the two L2 game sub-corpora, suggesting that the lexical complexity of the L1 game language had a distinct profile from that of the L2 games with localization.

Fig. 4
figure 4

Box plot of entropy by game.

Full size image
Table 2 Descriptive statistics of entropy for the three sub-corpora.
Full size table
Table 3 Tukey’s post-hoc test results of the entropy difference among the three sub-corpora.
Full size table

Comparing the MDD of the three sub-corpora

The MDD values for the texts in the three sub-corpora were calculated and the results are visualized in Fig. 5. A detailed statistical summary of these results is shown in Table 4. The MDD values of the three sub-corpora were similar, with minor differences: Sekiro had the highest mean value (M = 1.727, SD = 0.151), followed by Black_Myth (M = 1.716, SD = 0.164) and then RDR2 (M = 1.445, SD = 0.156). As when comparing the entropy values among the three sub-corpora, the assumptions of a normal distribution and homogeneity of variances were validated (p > 0.05), as shown in Table 4. The ANOVA test result (F = 226.730, p < 0.05) indicated a significant difference between the mean entropy values of the three sub-corpora. The findings of Tukey’s post-hoc tests are summarized in Table 5 and indicate that the MDD of RDR2 significantly differed from those of Black_Myth (p < 0.05) and Sekiro (p < 0.05), but there was no significant difference between Black_Myth and Sekiro. Moreover, the differences in MDD value between the L1 game sub-corpus (RDR2) and each L2 game sub-corpus (Black_Myth and Sekiro) were larger than the difference between the two L2 game sub-corpora, suggesting that the syntactic complexity of the L1 game language had a distinct profile from that of the L2 games with localization.

Fig. 5
figure 5

Box plot of MDD by game.

Full size image
Table 4 Descriptive statistics of MDD for the three sub-corpora.
Full size table
Table 5 Tukey’s post-hoc test results of the MDD difference among the three sub-corpora.
Full size table

Discussion

Simplification fails to capture localization

The findings of this study provide important insights into the linguistic complexity of localized video game texts and challenge prevailing assumptions within translation studies. Our analysis demonstrates that RDR2, representing an L1 game corpus, has distinct lexical and syntactic profiles compared with the L2 game localizations (Black_Myth and Sekiro). This distinction is evident in the significantly lower values of entropy and MDD metrics for RDR2 compared with its L2 counterparts. The entropy analysis revealed that the L2 localizations, Black_Myth and Sekiro, had a higher lexical diversity than RDR2. This finding contradicts expectations of lexical simplification (e.g., Lv and Liang, 2019) but supports the findings of Ferraresi et al. (2018), who documented greater lexical complexity in translated texts than their non-translated counterparts in certain contexts. Similarly, the MDD results indicated that the syntactic complexity levels of the L2 localizations were similar to or greater than that of the L1 corpus, challenging the claim of syntactic simplification (e.g., Xu and Liu, 2023) but aligning with Fan and Jiang’s (2019) finding that translated English texts had longer MDDs than native texts. Our results add complexity to the literature by directly contradicting the findings of studies that have tested the simplification hypothesis and found that translated texts were less complex than original texts in either the lexical or syntactic dimensions (Liu et al., 2022) or in both these dimensions (Fan et al., 2025). Our results suggest there is a limit of the implications to the simplification hypothesis and then the need for further research into other universals. The claim by Liu et al. (2022) that “the use of simplification as an umbrella term has failed to paint a genuine picture of a multi-faceted linguistic phenomenon” is echoed in the present study, the findings of which suggest that factors in localization, such as text type, genre conventions, and the specific demands of interactive media, significantly influence whether translation exhibits simplification tendencies. Consequently, translation studies must move beyond monolithic assumptions about universals and develop more nuanced frameworks that account for the diverse manifestations of translation across contexts and modalities.

Localization is not a mere translation

This complexification pattern observed in this study aligns with an emerging body of research that has identified exceptions to the simplification hypothesis, particularly in contexts in which the demands of localization go beyond mere translation. Video game localization, as noted by O’Hagan & Mangiron (2013), involves a comprehensive adaptation process that encompasses cultural, technical, and functional aspects.

Our results suggest that localization teams may inadvertently increase linguistic complexity rather than simplification in their efforts to create culturally resonant and contextually appropriate game experiences. Such complexification could be attributable to several factors. First, the need to adapt culturally specific references and maintain narrative depth might require more nuanced vocabulary and syntactic structures than those found in the source text. For example, there is some transliteration of character (“Baw-Li-Guhh-Baw”) In Black Myth, and place (“Hachiyouji Temple”) names in Sekiro. Also, Tie Shan Gong Zhu (Princess with an iron fan) is a specific character in Black Myth. By translating her name directly, the localization would maintain high cultural fidelity, preserving the original context and characteristics of the character. However, for an audience unfamiliar with Chinese mythology, this direct translation could increase cognitive load, making the character harder to understand. In contrast, the localization team used “Rakshasi,” a term from Hindu and Buddhist mythology, reduces cognitive load by providing a more familiar reference point. Yet, this choice sacrifices the specific cultural fidelity of “Tie Shan Gong Zhu,” altering her original cultural context.

Second, the interactive nature of video games demands text that can accommodate various player choices and gameplay scenarios, potentially leading to more diverse linguistic expressions. This requirement stems from a fundamental difference between video games and other media forms, which is that games are designed to engage users in dynamic, participatory experiences rather than passive consumption (O’Hagan, Minako, and Mangiron, 2013). For the process of localization, this interactivity introduces unique challenges beyond those presented in traditional translation. Localization is not merely the conversion of text from one language to another as it involves adapting the entire product to meet the linguistic, cultural, and technical expectations of different markets (Folaron, 2008). For video games, the localized text must remain coherent and engaging across various branching narratives and player-driven scenarios. The poetic clues for defeating bosses in Black Myth illustrate how localization teams must balance artistic intent with functional clarity; for example, the clue “Hit the back, let the beats respire, as in mist, thunders conspire” uses complex diction and rhyme to maintain the game’s atmospheric tone while providing necessary gameplay information. This approach reflects how localization must consider both the functional aspects of communication and also the aesthetic qualities that enhance player immersion. The localization process for such interactive elements requires specialized tools and quality control procedures to ensure functionality and linguistic accuracy (Esselink, 2000). Unlike traditional translation, which often deals with static text, game localization must account for variables, concatenation, and context-sensitive strings that may appear in different locations or combinations depending on player actions (O’Hagan et al., 2013).

Third, within the traditional dichotomy of translation strategies, gamers greatly prefer “foreignization” to “domestication” to preserve the feel and atmosphere of the original product (see Venuti 1995). This preference for foreignization may further contribute to the complexification observed in localized game texts, as localization teams strive to preserve the authentic experience while making it accessible to players in multiple cultural contexts.

Conclusion

This study provides novel insights into the linguistic complexity of localized video game texts and challenges the longstanding simplification hypothesis in translation studies. Through a quantitative comparison of lexical and syntactic complexity using entropy and MDD, we demonstrate that localized versions of two video games (Black Myth: Wukong and Sekiro: Shadows Die Twice) have a higher lexical diversity and comparable or greater syntactic complexity when compared with an original game text (Red Dead Redemption 2). These findings suggest that the multifaceted localization process may lead to complexification rather than simplification and are aligned with an emerging body of research highlighting exceptions to the simplification hypothesis, particularly in contexts in which localization demands go beyond mere translation.

Our findings carry important implications for the theoretical and practical aspects of translation and localization studies. For theory, the results underscore that supposed translation universals, specifically the simplification hypothesis, are not valid across all contexts. The complexification observed here in localized game texts directly contradicts the traditional assumption that translated texts are inherently simpler than original texts. For practice, our findings highlight the nature of localization as not a mere process of text conversion but rather a comprehensive adaptation that balances cultural authenticity, narrative coherence, and technical functionality. Localization teams must be aware that their adaptation choices can have a significant influence on linguistic complexity and the player experience. The need to adapt culturally specific references, maintain narrative depth, and accommodate interactive gameplay scenarios often requires more sophisticated linguistic choices, thus leading to complexification rather than simplification. This aligns with the growing recognition in localization research that localization demands more than mere translation (O’Hagan & Mangiron, 2013).

Our limitation is the focus on the English target language without source text, the restriction to dialogue, and the genre homogeneity. Future research should more deeply explore the factors influencing linguistic complexity in game localization, including genre-specific demands, cultural preferences, and the role of technology in the localization workflow. Also needed are cross-linguistic studies involving different language pairs and game genres to validate these findings and develop more nuanced frameworks for understanding translation phenomena in the digital media landscape. Ultimately, this study underscores the need for translation studies to move beyond monolithic assumptions about simplification and embrace the multifaceted nature of localization in modern digital contexts.