Units of Translation and the Limited Capacity of Working Memory

Abstract

A unit of translation is a source text fragment of any length or nature that piques a translator’s interest during translation. Alves and Vale proposed the concept of macro and micro translation units based on pauses and times identified in Translog software while translating a document. The objective of this exploratory study was to describe the characteristics of translation units in relation to the limited capacity of working memory (WM) and the information storage-processing capacity. Four trained translators participated in this study, translating a short text about underground mining. The data collection tools were Translog II and Camtasia software. Findings showed that participants used an analytic and sequential processing method and their limited capacity of WM exhibited a lower processing speed, which had an impact on the length of the macro units of translation (UTs).

1. Introduction

Translation is a long-standing human activity that allows individuals all over the world to engage, understand, and share their experiences. Alves and Vale [1] and Valero [2] states that translation is primarily “a fundamentally communicative activity” (p. 20–0). Hatim and Mason [3] point out that translation enables communication across linguistic and cultural boundaries. Translation, being a cognitively intricate endeavor, encompasses the creation of a target text (TT) within the parameters of production and reception. This process involves fulfilling its communicative function and facilitating text-to-text operations as well as requiring translators to engage their memory systems, attentional processes, and executive functions. By doing so, translators are able to comprehend the source text (ST) by delving into its significance and communicative intent, thereby ensuring a successful act of communication [4].

The late 1960s aroused a marked interest to better understand mental processes. This interest gave rise to conducting studies on aspects such as working memory (WM) and translation [5], the process of reading [6], the cognitive effort [7, 8, 9] and the distribution of visual attention [10, 11, 12]. All these studies have demonstrated the importance of further research on cognitive functions such as working memory due to its involvement in translation tasks. WM helps translators plan, reason, retrieve, retain information and make appropriate decisions. Therefore, empirical studies with keylogging, eye-tracking, and Event Related Potential (ERP) methodologies can contribute to expanding existing cognitive models and describing observable processes that account for translators’ behavior [13].

The interest in understanding how translators manage their WM resources emerged from a desire to fully comprehend how they process information throughout translation tasks [5, 14, 15]. This stage of processing also implied a segmentation in units of translation (UT) which have been addressed from different perspectives, including linguistics [16, 17, 18, 19, 20] and cognitive [21, 22, 23]. The former approached UTs from the product perspective. Results were focused on the perspective of the languages, whereas the latter considered the process, placing importance on establishing a link between WM and UTs [24].

The purpose of this chapter is to describe the characteristics of UTs with respect to the limited capacity of WM and its information storage-processing capacity throughout a translation task. According to Séleskovitch [25], Léderer [26], Gile [27], and Santamaría and Jiménez [28], the role of memory in translation tasks is of utmost importance. Conducting empirical studies “to account for the role of memory in translation tasks integrating translatology and cognitive psychology” (p.30) [5] may shed light on how translators improve their use of memory resources. This chapter contains four sections: first the theoretical foundations on which this proposal is based; second, the methodological procedure for accounting for units of translation and the limited capacity of working memory and information storage-processing capacity; third, the results and discussion sections about the main findings; and finally, the conclusions drawn from this exploratory study.

2. Theoretical framework

Currently, translators are compelled to adapt to the various working conditions that globalization has generated. Because of these new dynamics, they must be at the cutting edge of the new technological advancements and appropriately manage their internal, cognitive, and external resources to meet translation needs. This entails relying on declarative and procedural knowledge.

2.1 Cognitive translatology

Language constitutes an essential component of cognition, and interacts with various cognitive processes, including attention, memory, perception, and decision-making, among others [29]. The process of comprehending an ST and constructing its meaning necessitates the activation of these cognitive processes, as they collaboratively engage in this intricate interplay. As highlighted by Muñoz [30], the conveyed meanings in a translation task are not solely confined to the act of communication but also become imprinted in the minds of the communicators.

In recent years, the progress achieved in the field of cognitive sciences has made a substantial contribution to the development and application of cognitive approaches to translation. In this regard, Muñoz [31] has proposed the concept of cognitive translatology which aligns with third-generation paradigms advocating for a comprehensive evaluation of the interplay between mind, body, and context. Cognitive translatology’s fundamental goal is to provide a realistic, thorough, coherent, and cohesive account of a set of communication acts [30]. Furthermore, cognitive translatology aims to offer an interdisciplinary endeavor based on the 4EA cognition empirical approach of elucidating the cognitive processes underlying translation tasks. This means that cognitive processes should be regarded from an extended, embedded, embodied, enacted, and affective perspective.

According to Muñoz [32], cognitive processes are embodied because they are shaped by bodily experiences; they are embedded because the brain is integrated into the body and its environment; enactive because they are comprised of actions; extended because they rely on external tools to alleviate cognitive load; and affective because emotions influence behavior and some social activities require the ability to reason regarding the emotions of others. From this vantage point, translators derive meaning from their interactions within the context as well as their cultural, social, sensorimotor, and emotional experiences. In other words, the cognitive processes involved in complex tasks such as translation develop through interactions between the brain and the outside world. Similarly, in the 1990s, Risku [13] considered translation as a socio-culturally situated interaction in which meaning was generated through the interplay of the subject’s cognitive abilities and their immediate environment.

2.2 Units of translation

A unit of translation has been characterized from a cognitive standpoint based on the cognitive processes performed during a translation task. A unit of meaning, in terms of Séleskovitch and Léderer [23] is a fragment of information that interpreters contextually process to deliver a message.

Units of translation ne sont [...] nor mot pris isolément, nor the phrase définie grammaticalement comme sujet - prédicat, mais l’unité de sens, c’est-à-dire le segment de discours dont l’avancée à un moment donné fait prendre conscience à l’auditeur ou au lecteur du vouloir dire désigné par linguistique formulation (p. 268) [23].

This suggests that the unit of meaning transcends linguistic and grammatical boundaries by not being regarded as separated words or grammatical structures of a sentence (subject and predicate). This proposal allows for a link with the UT when cognitive components of speech segmentation are considered and evidences the importance of information processing and retrieving in interpreting tasks as well as in translation tasks.

From the translation perspective, Alves and Vale [1] defined UTs as segments of the ST of any length or nature that demand the attention of translators to produce a TT. They also suggested that these UTs be thought of as micro and macro units of translation in terms of time and pauses. A micro UT, according to them, is “the flow of continuous TT production- which may include the continuous reading of ST and TT segments- separated by pauses during the translation process, as registered by Key-logging and/or eye tracking software.” (P.129). A macro UT is as “a collection of micro UTs that includes all interim text productions that follows the translators’ focus on the same ST segments from the first tentative rendering to the final output that appears in the TT” (p.129).

In 2011, Alves and Vale [1] presented a taxonomy of macro units (P1, P2, and P3) in accordance with the adjustments made during the translation process phases described by Jakobsen [33], namely: 1) the beginning orientation phase; 2) the middle drafting phase, and 3) the end revision phase. These study findings demonstrate that adjustments in the macro units of type P1 occur during the drafting phase, adjustments of type P2 occur during the end revision phase, and adjustments of type P3 occur during both drafting and revision phases. As a supplement to this taxonomy, Alves et al. [34] added the concept of a macro unit of translation, P0 that does not undergo any adjustments during the prior phases described.

The units of translation have also been linked to WM functioning; Dragsted [35] suggested that the length and nature of such units from a cognitive perspective correspond to information segments processed during translation which are constrained by limited capacity of WM. She further claimed that while cognitive scientists disagree on the precise capacity of WM, it is generally agreed that it can only hold 3–7 information items. Furthermore, the amount of WM involved in the UT segmentation is determined by the level of ST difficulty.

On this basis, difficulties in the text can be expected to have various types of impact on the length and nature of UTs and on the speed of production. As Newell and Simon [36] and Campbell [37] argue, difficult items in the source text place a great demand on the limited capacity of working memory. When a large amount of the total WM capacity is used to comprehend or produce a particular item, it must be assumed that less, or no, capacity is left to concentrate on other elements, and that consequently the presence of a problematic lexical item in the ST will reduce the number of items in a UT, possibly to only one word [35]. In the current study, the UT is defined as an information segment that undergoes cognitive processes during a translation task.

Furthermore, according to Dragsted [24], pauses are important since they can reflect cognitive processes and demarcate boundaries between text production and units of translation. As a result, WM plays a pivotal role in both comprehending the ST and producing the TT, as well as in delimiting the processed segment of information. Pauses and TT adjustments are made to determine macro and micro UTs, and these pauses provide useful information for decision making and solving translation problems.

2.3 Working memory and its operating characteristics

Working memory (WM) has raised significant attention in translation studies as an implicit component of the translation process [5, 6, 24, 35, 38, 39, 40], as translation encompasses various cognitive processes [41]. According to Dragsted [24, 35], WM is critical in the segmentation of a text in UT when comprehending a text for translation purposes. This is consistent with Macizo and Bajo [6] who demonstrated that WM is not only a vital cognitive prerequisite for comprehension in general, but also necessary for the simultaneous activation of two languages and their switching.

Dragsted [24] asserts that when translators are processing new segments, they are unable to retain previously processed segments in WM. Consequently, they tend to forget previously proposed solutions and occasionally need to revisit the translated text to recollect them. As previously mentioned, the translation process requires a substantial amount of WM resources, encompassing parallel activation of two languages, comprehension, and production of the target language (TL), and TT, as well as the segmentation of the text into micro and macro UTs.

This chapter has adopted the working memory model proposed by Cowan [42, 43] and Cowan et al. [44] for the purposes of discussion. Cowan’s model provides an integrated perspective that highlights the interconnectedness of memory and attention processes. While incorporating certain components from other WM models. This model introduces an embedded framework and emphasizes the crucial role of attention in stimulus processing. This cognitive process holds great significance as it regulates the activation, retention, and manipulation of representations within WM [45].

According to Cowan et al. [44], WM is “the ensemble of components of the mind that hold a limited amount of information temporarily in a heightened state of availability for use in ongoing information processing (p. 45).” Cárcamo [46] provides an explanation of Cowan’s model, which states that long-term memory serves as a reservoir of information, while WM encompasses two levels of activation within this long-term memory that are independent of modality. In other words, there are no specific modules for each type of stimulus.

The first phase of WM is associated with motor persistence, while the second phase pertains to storage, conceptualized as a memory of the stimulus involved. In the later phase, attention is employed to activate a set of features that align with the stimulus in long-term memory. Within the activated segment of long-term memory, a specific subgroup of items present in the focus of attention is also activated. This process is initiated by the registration of sensory information and regulated by the central executive, as illustrated and explained by Cowan et al. [44] (see Figure 1).

Figure 1.

Embedded-processes model. (1) habituated stimulus, (2) physically changed stimulus recruiting attention and orientation, (3) deliberately attended stimulus, (4) information deliberately retrieved from long-term memory, and (5) automatic association that attracts attention source: Cowan et al. (p. 64) [44].

An essential aspect to consider is the limited capacity of the focus of attention. Miller [47] introduced the concept of the “magical number seven plus or minus two,” suggesting that individuals can actively maintain and efficiently process around seven items or chunks in their working memory, with a slight variation of plus or minus two. However, Gilchrist et al. [48] argued that people can recall only three or four verbal chunks. Other authors have expressed uncertainty regarding the exact limit of WM capacity, proposing that it may depend on specific intricacies of the memory task [49].

Cowan [49] asserts that the capacity of WM exhibits notable variability depending on the specific processes employed in a given task. For instance, when individuals need to memorize verbal material, they may employ strategies such as rehearsal, mentally repeating the information or they may create chunks by grouping multiple words together. Similarly, when memorizing a sequence of spatial locations, individuals can use mental imagery to envision a pathway composed of these locations. The author also states that:

Tests of working memory demonstrate practical limits that vary, depending on whether the test circumstances allow processes such as grouping or rehearsal, focusing of attention on just the material relevant to the task, and the use of modality- or material-specific stores to supplement a central store. Recent work suggests, nevertheless, that there is an underlying limit on a central component of working memory, typically 3–5 chunks in young adults. [...] (p. 5) [49].

Within this chapter, Cowan’s model serves as a valuable theoretical framework to investigate and analyze both the macro and micro units of translation identified in the translation task of a specialized text. The analysis focuses on examining macro UTs observed within the Translog II software, Camtasia screen recordings and semi-structured retrospective interview. By using this model, this study provides intriguing insights into the segmentation strategies employed by the participating translators when working with specialized texts.

3. Method

The data collected for this exploratory case study of descriptive scope was primarily quantitative in nature with the purpose of establishing a relationship between the UT and the WM functioning features through the analysis of quantitative and categorical variables. Although this is an exploratory study comprising four cases, it does not lead to generalizations, it does provide an interesting viewpoint on the cognitive processes involved in the UT segmentation and WM’s limited capacity during a translation task.

3.1 Participants

Four certified female translators from different regions of Colombia voluntarily agreed to participate in this case study. They all stated that they had no visual or neurological condition that prevented them from participating. Holding an undergraduate or graduate academic diploma in translation awarded from a national or international university, Spanish as a native language, one year experience in English-Spanish language combinations were considered as inclusion criteria. Their ages ranged between 29 and 36 years old with one of them having lived in an English-speaking country. They all signed an informed consent and provided sociodemographic information.

3.2 Materials

The software programs Translog II and Camtasia were used. The first software logged all keyboard activities in a systematic manner, whereas the second created a screen registration during the translation assignment’s execution. A semi-structured retrospective interview was also conducted, as proposed by Ericsson and Simon [50], who urged doing so after finishing an assignment. The interview covered topics such as segmentation, planning, revision, dual language processing, and repetition. The validity of this instrument was determined by expert opinion.

3.3 Procedure

Participants first signed the informed consent form and completed a demographic survey. Second, all translators were asked to translate a 233-word text from a scientific journal in the field of environmental sciences, a sub-domain of climate change. This was done with the Translog II software program, and the entire process was recorded with Camtasia software. There was no time limit in executing the translation task. Third, the semi-structured retrospective interview was conducted right after each participant completed the translation task.

4. Results

To conduct the data analysis, the first step involved organizing the data gathered by using the Translog II software that served for the identification of the macro UTs, creating a written description of the Camtasia screen recording and transcribing the interviews’ information. Subsequently, the macro UTs of translation were identified.¹

The macro UTs were analyzed according to the taxonomy proposed by Alves et al. [34]. The number of macro UTs (P0, P1, P2, and P3), and the word count for each segment were identified. As for the translation process, the phases established by Jakobsen [33] served for the identification of UTs according to the taxonomy previously mentioned. All this was contrasted with the interview and the description of the Camtasia screen recordings.

Finally, the data was examined in light of the characteristics of UTs in relation to the limited capacity of WM and its role in information storage-processing. as suggested by Andréu [51], the interview and Camtasia data enable “inferences or explanations in a given reality through communicative messages” (p. 9), however, in this particular case, they were subjected to quantitative analysis.

To establish the pause value, although there was no consensus regarding the optimal length of a pause, a duration of two seconds was determined as indicated by Dragsted [24]. Similarly, according to Cowan et al. [44], in activated long-term memory, the storage process involves attention-based refreshing and verbal rehearsal for maintenance. Decay of poorly encoded information may be associated with memory loss over time, with the encoding process as well as the familiarity with the items that are required to be retrieved. Thus, Translog II software was configured with this pause value, and a linear record of each participant was extracted (see Figure 2).

Figure 2.

Translog software linear view. Each dot corresponds to a 2-second pause. The left arrow indicates eliminations, while the up and down arrows represent clicks. Source: Authors’ elaboration.

Table 1 displays the macro UTs for each participant. It is evident that the most prevalent macro UT types among all participants were P0 and P1, with 206 and 86 instances, respectively. In contrast there were only 21 instances of macro UT type 2 and right instances of macro UT type 3.

Participant	P0 (No modification)	P1 (drafting phase)	P2 (end revision phase)	P3 (changes in both phases)	Total macro UTs
T1	47	6	6	2	61
T2	52	12	10	3	77
T3	65	38	0	1	104
T4	42	30	5	2	79
TOTAL	206	86	21	8	321
MACRO TUs according to the taxonomy

Table 1.

Macro translation units for each participant.

Source: Authors’ Elaboration. Data obtained from Translog II software.

According to Alves et al. [34], an additional means of measuring task difficulty and cognitive effort invested during a translation task is through the number of macro UTs. Among these macro UTs, it is noted that type P0 exhibits a lower level of difficulty compared to macro UTs P1, P2, and P3. These findings suggest that despite this translation task involving a text from a specific domain of knowledge with intricate grammatical and lexical structures, all participants showed a greater frequency of macro UTs classified as type P0.

However, these findings reveal that two participants (T3 and T4) exhibited higher frequency of macro UTs classified as P1 compared to the two other participants (T1 and T2). This discrepancy resulted in a reduction in the number of macro UTs categorized as P2 and P3 during the revision phase, indicating that a significant portion of the comprehension and monitoring process occurred during the TT production phase. Notably, T3 and T4 initiated the revision phase at 8229.926 ms. accounting for almost 99% of the entire process, while T4 started this phase at 5452.797 ms, accounting for 93% of the entire process. Furthermore, T3 and T4 exhibited the lowest values during the revision phase.

These findings also reveal that all participants’ behavior was different regarding the macro UTs categorized as type P3 that, according to Alves et al. [34], is “where processing effort seems to be concentrated.” This indicates that both the segmentation process and the level of task difficulty were in different parts of the text (see Figure 3).

Figure 3.

Macro UTs of type 3. Source: Authors’ elaboration.

Similarly, the ST segments identified by the participants as particularly challenging to translate do not entirely correspond to the macro TUs categorized as P3, which basically implies a high level of complexity in comprehension and production. This coincides with Alves and Vale [52] who assert that “P2 types are more frequent than P3 types and more substantial revisions are only found among P2 types of macro translation units. P3 types seem to account for more fine-grained revisions which are quite small in numbers.” (p.120), which means greater WM capacity.

The data provided by Translog II also allowed for concluding that the macro UTs P1, P2 and P3, which contain modifications in different parts of the translation process and therefore are indicators of the focus of attention of the translator, were formed mainly by terms. This indicates that, despite the high number of macro UTs P0, the TT contained complex units for the translators and that the changes presented in the UTs were not simply related to modifications of orthographic or coherence type, but there were variations of pragmatic and semantic type.

With respect to the limited capacity of WM, there is still a lack of consensus between the exact amount of information processed by the WM, as previously mentioned. The number of elements kept in the WM was once thought to be seven in the early stages of Miller’s idea. However, according to Dragsted [24], “most investigators now tend to agree that the number seven is too high, and that the maximum amount of information items is closer to four or less” (p. 44), such as the case of Gilchrist et al. [48] who identified four items to limit the WM capacity. These findings reveal that for two of the participants the most frequent length segment was one word (T2 and T3); three words for one participant (T3) and two words for one participant (T3). To a lesser extent, there were UTs ranging between four and five words. Figure 4 exhibits the length of UTs per participant.

Figure 4.

Length of units of translation. Source: Authors’ elaboration.

The analysis of the Camtasia recordings and interviews showed that the participants primarily conducted searches for challenging segments and terminological units to translate. This observation corresponds to Wilss’ [53] proposition, which emphasizes the impact of a translator’s skills, abilities, and experience on the information activated and processed during a task. All participants acknowledged their limited experience in translating specialized discourse, which consequently heightened their cognitive load in completing the task.

With respect to WM information storage-processing, Gathercole et al. [54], Baddeley [55], Just and Carpenter [56], as cited in [57], and Cowan et al. [44] assert that WM is currently understood as “a brain system that provides temporary storage and manipulation of the information necessary for complex cognitive tasks, such as language comprehension, learning, and reasoning” (p.25). Based on this assertion and considering the prevalent length of the macro UTs, these findings suggest that the capacity of information processing and storage may have influenced the text segmentation into UTs. It appears that participants did not retain large amounts of information to complete the translation task. Instead, significant pauses were observed during the comprehension of relatively smaller segments.

As for the source text and target text processing, it is noted that translation is a complex cognitive task that involves working with two languages and cultures, necessitating the simultaneous use of two lexicons by translators. As highlighted by Alves and Hurtado [58], Alves and Vale [52], and Alves et al. [34], pauses observed during a translation task provide valuable insights into the cognitive processes at play. These pauses signify periods in which translators engage in a range of cognitive activities related to executive functions, including comprehension, planning, revision, monitoring, and problem-solving. The findings indicate that three out of the four participants devoted significant time to the initial orientation phase including activities such as documentation and terminology searching. However, one participant exhibited a different behavior pattern. Whereas the middle drafting phase exhibited the activation of executive functions as evidenced by highest average pauses during the production of the TT (see Figure 5).

Figure 5.

Translation process phases.

The notable presence of extended pauses indicates that all participants engaged in an analytical processing approach of “first analyzing and fully comprehending the SL segment, before moving to produce the TL segment” (p. 150) [24]. This type of processing requires additional time because of the documentation process. Analysis of the Camtasia screen recordings revealed that participants mainly relied on websites such as Linguee, glossaries in English and Spanish, books in PDF format, web pages, videos, and images as search and documentation sources. This behavior aligns with the functions of the central executive which entails supervising, transforming, and cognitively manipulating the encoded information. In particular, T1, T2, and T4 participants mentioned that they segmented the text into units of meaning, reflecting their cognitive involvement during the translation task.

This indicates that each pause was directly related to the ability of the subjects to understand each segment, so they used the necessary time before starting the translation of each segment. It can be evidenced then that the UTs are conceived as a unit of meaning that transcends the linguistic and grammatical level by not being considered isolated words [59], but rather takes into account cognitive aspects during the segmentation of discourse. It is not a question, then, of translating word-for-word [4].

5. Discussion

Research conducted on different types of UT and WM during the translation process have provided valuable insights into how translators comprehend source text information and produce a target text. These contributions have been crucial in understanding text segmentation in terms of suppression, selection, inhibition, long-term memory activation, and coordination. However, there is still a need for further exploration of how translators activate cognitive processes – particularly WM – to make appropriate adjustments in relation to macro and micro UT.

These findings underscore the role of WM in the process of segmenting a specialized text. This section addresses the topic from two main perspectives: the WM model as proposed by Cowan [42, 43] and Cowan et al. [44], and the framework of Macro and Micro UTs introduced by Alves and Vale [1] and expanded upon by Alves et al. [34]. These theoretical perspectives provide valuable insights into the cognitive processes involved in the segmentation of specialized texts, highlighting the role of WM in this complex task.

In relation to Cowan’s model, the authors of this study emphasize the role of working memory and attention, as well as the interplay between these cognitive processes. They also highlight how attention plays a crucial role in activating specific information from long-term memory during the segmentation of the text into UTs. Both WM and attention facilitate the retrieval and activation of relevant linguistic and contextual knowledge necessary for effective text segmentation during the translation process.

In the present study, participants were observed accessing long-term storage systems – particularly, lexical memory and semantic memory storage systems – to carry out text segmentation. The ability to access these memory systems relies on the capacity to exercise control over the focus of attention which may manifest as either cognitive flexibility (change of focus to choose another information option) or cognitive stability (permanence of focus to analyze the chosen information). In Cowan’s [42, 43] and Cowan et al.’s [44] terms, this cognitive control, which is critical in managing attention and long-term memory information retrieval, refers to the central executive level.

Cowan [49] also states that depending on the task, individuals employ different strategies to recall and retrieve information which can typically involve three or four chunks of information. However, findings of the present study revealed that the most common quantity among all participants ranged between one and three chunks (words) with a predominance of one chunk (one word). This pattern may be attributed to the challenges posed by specialized texts, which can make the segmentation process more demanding for translators. This finding is consistent with Dragsted’s [35] assertion that a difficult lexical item in the source text can reduce the length of a UT to a single chunk. Furthermore, this observation aligns with the researchers’ perspective who argue that the limited capacity of WM is influenced by task difficulty, as translation involves working with two cultures, the parallel activation of two languages, and their switching [6], in order to provide a realistic, coherent and cohesive account of a set of communicative acts, as proposed by the cognitive translatology [30].

From a translation perspective, the findings of this study suggest that translators direct their attention toward lexical and semantic elements that facilitate text comprehension, with a specific focus on micro and macro UTs. Attention was employed to selectively attend to and filter necessary information that was processed by WM, thus enhancing text comprehension for completing the translation task. The observed segmentation patterns, characterized by the utilization of one to three chunks, suggest that the specialized field of knowledge presented challenges for the participants. Consequently, they were unable to retain previously processed segments and had to refer back to the translated text for recollection, which aligns with the findings of Dragsted [24].

Figure 6 provides a comprehensive overview of the segmentation process in terms of macro and micro UTs during the orientation phase, leading to the production of a TT. This representation incorporates elements from both the Cowan WM model and the taxonomy of UTs proposed by Alves and Vale [1] and Alves et al. [34]. It offers a plausible framework to elucidate the dynamics of text segmentation during the translation process, taking into account the WM operational characteristics.

Figure 6.

Integrated representation between working memory model and macro UTs. Source: The figure presented in this study was developed by the authors, drawing upon the concepts and frameworks of the embedded-processes model proposed by Cowan et al. [44], as well as the taxonomy of translation units. The taxonomy includes the classification of macrounits: P1 for adjustments made during the drafting phase, P2 for adjustments made during the end revision phase, and P3 for adjustments made during both the drafting and end revision phases [1]. Additionally, P0 represents macrounits that do not undergo any adjustments [34].

This representation suggests that the WM is a center for processing and storing information, and translation is the product of the interaction of different controlled and intuitive processes, as mentioned by Kiraly [60]. During the segmentation of information in UTs, translators circumvent numerous situations that require cognitive efforts such as remembering information, understanding terms, finding solutions to lexical or terminological problems, maintaining attention, and reviewing. All these efforts rely on WM functioning and the central executive deliberately activates long term memory storage to suppress, select, coordinate, and inhibit.

6. Conclusions

The limited capacity of WM had a significant impact on the characteristics of the UTs. The most frequent segment length among all participants ranged between one and three words, one word being the most predominant UT accounting for 50% of the total number of macro UTs. This finding suggests that all participants processed smaller units because of the particularities of the task, requiring them to focus their attention on concrete segments of the text, such as terminological units. This cognitive demand is evidenced in the pauses observed during the orientation and writing phases, reflecting a decrease in the processing speed.

The information was segmented mainly at the word level, and there was evidence of long pauses between the UTs associated with macro planning activities. Translators considered the UTs as units of meaning, which enhances the involvement of cognitive processes during information processing in translation tasks. This perception suggests a non-linear understanding of information resulting in frequent searches for completing the task.

The activation of various components of WM including supervision, long-term memory activation, coordination, updating, problem solving, inhibition, and self-monitoring play a significant role in text segmentation into UTs with respect to task complexity. All these components are crucial in translation tasks, and in particular, when dealing with specialized discourse.

Acknowledgments

The authors would like to express their sincere gratitude to Yurany Muñoz Santanilla for providing the data for this proposal, and to Gregory Wallace Amos for revising this final chapter.

Conflict of interest

The authors declare no conflict of interest.