Many roads lead to Rome: an empirical study of summarizing translation processes

3.1 Data collection

In the academic year 2021–2022, 35 third-year undergraduate students studying Applied Linguistics (University of Anon) who were enrolled in the Translation course Spanish > Dutch 2 were asked to produce a summarizing translation of an article by Manuel Villoria Mendieta (2021) titled Rendir cuentas para transformar y renovar España, published in the Spanish newspaper El País (see Appendix I). The central topic or main theme of this article is how well Spain scores in the quality of its democracy and which measures can and need to be taken to improve this quality. This ST contained 1,170 words, which had to be reduced to 500 in the TT for publication in a Dutch newspaper. The students were provided with an electronic and a paper copy of the ST; they were allowed to make notes on the ST handout. They could work on this task in class for 3 hours, using MS Word and any digital sources and tools of their choosing, except MT systems, social media and e-mail. These changes were introduced to create a working environment that resembled the normal classroom setting (i.e., no MT) and students’ personal preferences (i.e., a paper copy of the ST). These students had already produced several other summarizing translations in this translation course and general summaries in their language courses. To increase the motivation for this particular task, individual product and process feedback was given after wards.

The translation processes of the students were registered using the unobtrusive computer keystroke logging software Inputlog (Leijten and Van Waes 2013). The course teacher (i.e., the author of this paper) also observed and noted down the students’ interactions with the paper copies of the ST. All the paper copies of the ST, including the students’ names, were collected afterwards.

3.2 Data analysis

The level of data attrition, owing to technical challenges, was 26 %, which led to a total sample of 26 students. The data analysis focused on process data. For the analysis of the keystroke logging data, both the general and source analysis files and the process graphs that Inputlog automatically generates were used. The general analysis file gave insight into what the students had typed, when and how long they paused, what and when they revised, and when they consulted external sources, among other activities. The source analysis provided information on the external sources that the students used, when and how long they used them, and also when and how frequently they switched between different windows (i.e., of the ST, the TT, and external sources). The process graphs yielded an insightful visual impression of the evolving text-production process, that is, the interaction between the number of keystrokes produced during the process and the number of keystrokes that the product contains, text movements, and consultation of sources outside the TT environment. The general file and process graph, in combination with an analysis of markers on the ST paper copies (e.g., handwritten notes, schematic overviews, arrows) and the researcher’s observational notes constituted the primary sources with which to identify strategies and, if possible, certain process commonalities between students. A qualitative process analysis of each participant was carried out using the annotation scheme presented in Table 1.

4.1 Summarization taking place during the pre-writing phase

The first phase of the translation process is called the pre-writing stage (Jääskeläinen 1999), the pre-writing phase (Englund-Dimitrova 2005), or the initial orientation phase (Jakobsen 2003). Englund Dimitrova (2005: 86) provides the following definition of the pre-writing phase: “[it] begins when the participant has received the ST and the oral information about the translation brief, and finishes when the participant starts to write down the TT as an integral text. Making notes about word meanings etc. while reading the ST for the first time is not considered as a start of the writing phase.” In the present study, the average duration of the pre-writing phase was 13.38 % [range: 0.98–65.32 min] of the total process time.

Although the students were provided with an electronic and a paper copy of the ST, most of them used the latter, on which they highlighted sentences, phrases, and words and made notes. These activities were not registered by Inputlog but could be reconstructed through the triangulation of the researcher’s observational notes, analysis of the handouts, and instances in the keystroke logging data where “nothing” happened. As shown in Table 2, the logging data, paper copies of the ST, and observational notes of 7 out of the 26 students revealed no discernible indications of weighing ST content units or the creation of a discourse hierarchy during the pre-writing phase. In contrast, the other 19 students displayed behaviour that could indicate relevant content selection and/or hierarchical discourse organization, sometimes even combining different strategies.

Table 2:

Frequency of content weighing strategies and hierarchical discourse strategies in the pre-writing phase.

Of these 19 students, 11 developed a schematic outline of the content units that should be featured in the TT, to varying degrees of detail (see Figure 1 for an example, in which the student anticipates 7 paragraphs in the Dutch TT, almost all of them with a number of talking points). Ten of these 11 students drafted this schematic outline in the target language.

Figure 1:

TT outline drafted in the pre-writing phase.

Of these 11 students who drafted a schematic outline, three developed the content outline without highlighting ST elements while reading the ST, whereas eight others adopted this planning strategy after having highlighted ST keywords and phrases (four students; see Figure 2 for an example) or entire ST sentences (one student; see Figure 3 for an example).

Figure 2:

Highlighting of ST keywords and phrases in the pre-writing phase.

Figure 3:

Highlighting of ST sentences in the pre-writing phase.

Three students employed a combination of strategies: they highlighted keywords, phrases (including discourse markers), and entire sentences in the ST, and manually recorded the main ideas of each paragraph in the ST margin, followed by drafting a concise schematic overview of the TT content in MS Word (see Figure 4 for an example).

Figure 4:

Highlighting ST keywords, phrases, and sentences in the pre-writing phase.

The eight students who did not draft a TT outline performed semantic reduction by highlighting elements in the ST. While the systematic highlighting of entire sentences was slightly more common (three students) than that of keywords and phrases (two students), there were also instances (three students) where both strategies were employed.

Whereas the highlighting of keywords and phrases suggests that salient ST content-detection and weighing took place, it cannot be ruled out, based on the data-collection methods used, that those students who highlighted sentences or words, phrases, and sentences actually used the highlighting strategy as a means of comprehending the ST content in depth. Yet the use of different colours for highlighting shown in Figure 4 might indicate both content weighing and hierarchical discourse organization: the discourse markers are underlined in blue, the (numerical) details are underlined in green, the main points are signalled by a red line, and there is also a consistent use of purple (enumerations). The pre-writing phases of “full translation” processes may entail this strategy, but do not generally do so. The same is true for the TT outline strategy, yet it is a typical planning strategy observed in writing research. Previous research by Galbraith et al. (2005) and Kellogg (1988, 1994) has suggested that drafting an outline before writing enables writers to execute higher-level problem-solving more effectively than doing so without pre-planning; and that it is consequently typically associated with the production of higher-quality texts. As is shown in Section 4.2, these ST markings and TT outlines appear to have served as cognitive artefacts, “physical objects made by humans for the purpose of aiding, enhancing, or improving cognition” (Hutchins 1999: 126). They help to perform the cognitive task of TT production, serving as a memory cue and/or a guide as to which content to integrate into the TT.

4.2 Summarization taking place during the writing phase

The writing phase of the translation process “begins when the participant starts to write down the TT and finishes when (s)he has written down an integral version of it” (Englund Dimitrova 2005: 86). As shown in Table 3, a minority of the students – 11 out of 26 – used a single TT production strategy consistently throughout the writing phase. However, most of the students combined several strategies for TT production, even though they did have a specific strategy as a starting point (e.g., translating the first ST paragraphs in full or using the TT outline they had previously drafted).

Table 3:

Frequency of drafting strategies in the writing phase.

The students who used a single TT production strategy differed in the strategy they opted for: some used literal patchwork translation, while others used paraphrasing. The literal patchwork translation strategy entails employing a default literal translation strategy for ST sentences and/or specific words and for phrases from one or multiple ST sentences and combining these to form a coherent TT sentence. The translation and integration of ST units is generally approached in a linear manner, following the order of ST sentence(s). Seldom are units from different ST paragraphs combined in a single TT sentence or paragraph. As shown in Figure 5, this literal patchwork strategy implies omitting certain words and/or phrases from the ST sentence. These are represented by the units crossed out with the red line in the English gloss of the Spanish ST excerpt in Figure 5. The remaining ST words and phrases are then translated in a literal manner. Occasionally, additional semantic reduction by way of deletion can be observed when the students revised the TT sentence they had just produced, as is the case in Figure 5 with “and how they are met”, and/or when they revised the TT paragraph they had just drafted, as illustrated by the revision represented by the purple line.

Figure 5:

Example of the literal patchwork strategy.

The words, phrases, and sentences that were translated literally were in almost all cases those that the students had highlighted if they had used the highlighting strategy during the pre-writing phase. This suggests that there was an interplay between the planning and the TT production strategies. Only four out of the 16 students who had used the highlighting strategy translated additional ST items that they had not highlighted before.

The other frequently employed TT production strategy was paraphrasing: here, the translator verbalizes the ST in their own words in the target language. This strategy implies not only translating the ST content less literally but also integrating content from the ST units in a less linear way, jumping back and forth in the same ST paragraph or even between ST paragraphs.

As illustrated by the example in Figure 6, summarization via paraphrasing may encompass several strategies:

1. the omission of ST units, as represented by the red lines;

2. linguistic reduction via generalization, as occurs, for example, with “this must however be done”, instead of the literal translation “must be renewed or withholdable”, or “their practices” versus “the ways in which it has accomplished them”;

3. linguistic expansion, as can, for example, be observed in “the Spanish citizens have every interest in being informed” instead of the more literal option, “the importance of”;

4. construction, which is described by Shreve (2006) as creating new macro-propositions from longer sequences of propositions. This is exemplified by “elections are considered as an obvious and appropriate way to render account for those who have taken up a political office”, which is constructed based on “the accountability implied by elections”, “representative democracy”, and “election is presented as the obvious method”.

Figure 6:

Example of the paraphrasing strategy.

The paraphrasing strategy often involves a more pronounced focus on TT cohesion. In the example shown in Figure 6, the translation student in question established cohesion with the final sentence of the preceding TT paragraph, which had emphasized the necessity of implementing control mechanisms that strengthen rather than paralyse government. Moreover, this immediate foregrounding of elections helped her to ease into an explanation of the accountability model proposed by the ST author. This highlighted the need for accurate information about the government’s feats, albeit without explicitly stating that the ST author has some ideas for improving the political accountability methods (as indicated in the first sentence of this ST paragraph).

The paraphrasing strategy, especially when employed in combination with the outline planning strategy, might to a certain extent resemble the knowledge-constituting strategy found in writing research, whereas the literal patchwork strategy might hint at a knowledge-telling strategy (Bereiter and Scardamalia 1987). As explained by Baaijen and Galbraith (2018): “Knowledge-telling involves the linear production of ideas in the order that they are retrieved from memory and the spontaneous translation of these ideas into words. Knowledge-transforming, by contrast, involves modifying the order in which ideas are produced, and deliberating over how to formulate these ideas to satisfy rhetorical goals”. In the present study, the highlighted ST units appear to have served as memory cues; and when the literal patchwork translation strategy was adopted, the ideas represented in these units were produced linearly and without semantic transformation, resembling knowledge-telling. In contrast, those students who started from a TT outline and paraphrased the (previously highlighted) ST units seemed to evaluate, reorganize, and rephrase these ideas to satisfy rhetorical goals, as writers whose approach is characterized as following a knowledge-constituting strategy do. The decision to adopt a literal patchwork translation strategy or a paraphrasing strategy might have been influenced by the way the translation students perceive summarizing translation, how static or dynamic their concept of translation is, and/or the way they conceptualize their role as translator (Chodkiewicz 2020).

A comparison of the pausing behaviour pertaining to the examples above provides some insights into how cognitively demanding each strategy might be. As shown in Figure 7, the TT production by the student employing the literal patchwork strategy is characterized by frequent short and longer pauses (30 in total; predefined pause threshold of 1,000 ms), the latter primarily preceding clauses and sentences and the former preceding lower-level units such as phrases and words, or following them in the case of typos. This is a typical pausing pattern previously found, for example, by Immonen and Mäkisalo (2010) and Puerini (2023). The student employing the paraphrasing strategy follows the same pattern but pauses much more frequently (74 times in total) and the number of pauses of more than 10,000 ms (highlighted in red in Figure 7) are more frequent too. Although the TT production of the student employing the literal patchwork translation strategy is not completely linear due to the many typo corrections, the non-linearity of the drafting process of the other student is more substantial: there is more iterative online revision within sentences. These indicators seem to suggest that the paraphrasing strategy is more cognitively demanding than the literal patchwork strategy. However, more in-depth quantitative analysis of pauses and text-production bursts is necessary to establish whether this is indeed a characteristic difference between the two strategies or rather idiosyncratic of these two translation students.

Figure 7:

Pausing in keystroke logging data of literal patchwork translation strategy and paraphrasing strategy.

Not all of the students immediately summarized the ST content in their writing phase. In fact, the TT production strategy typically found in translation process research, in which translators translate entire ST paragraphs in a linear fashion – starting with the first paragraph, then the second, the third and so on – could be observed partially in the writing phase of four students as they translated all of the sentences of several ST paragraphs. It appears that these students, after having employed this strategy for several ST paragraphs (the first 2, 4, 6, and 7 paragraphs, respectively), recognized its time-intensive nature, which prompted them to change their strategy. The rationale behind this shift cannot be inferred from the available data, but the increasing semantic complexity of subsequent ST paragraphs and/or the dwindling task time might have been influential factors.

During the writing phase, nine students used the schematic TT content overview as a starting point for their TT production. Seven either “filled in” the TT outline or initiated TT production based on the outline by translating previously highlighted words, phrases, and/or sentences in a predominantly literal manner (three students) or through paraphrasing (three students). Two other students adopted a default literal translation approach for the ST units that they had highlighted in the ST and their concise TT outline, but also applied the same strategy for other previously unselected ST elements. The student who had not highlighted any ST units during their ST reading but had developed a detailed schematic overview of the ST ideas that should be included produced the TT based on this outline. Then, during the TT production, they transformed the outline items into full sentences while omitting certain content points. Since these TT sentences correspond to a literal translation of certain ST sentences, it is likely that during this verbalization stage this student also used the ST, although this cannot be verified using the available data.

From the previous results it can be deduced that there is a certain interaction between the planning and the TT production strategies. As shown in Table 4, 12 of the 16 students who had used the highlighting strategy in the pre-writing phase translated the highlighted ST units during the writing phase. However, no clear preference could be observed for the literal patchwork strategy or the paraphrasing strategy among these students who had shown signs of content weighing and/or hierarchical discourse organization in the pre-writing phase. Four of them also translated additional ST items. These four differed in the type of ST units they had highlighted. Moreover, three of them translated the previously highlighted ST units in a literal patchwork manner, whereas the other preferred the paraphrasing strategy.

Table 4:

Frequency of planning and TT production strategy combinations.

Planning	TT production	No. of participants
Highlighting keywords and phrases in ST	Literal and linear translation of previously highlighted keywords and phrases + literal and linear translation of a selection of ST phrases and sentences
Highlighting keywords and phrases in ST	Paraphrasing translation of previously highlighted keywords and phrases
Highlighting key sentences in ST	Literal and linear translation of previously highlighted sentences
Highlighting key sentences in ST	Paraphrasing translation of previously highlighted sentences
Highlighting key sentences in ST	Paraphrasing translation of previously highlighted sentences + paraphrasing translation of a selection of ST phrases and sentences
Highlighting keywords, phrases + sentences	“Full” translation of ST paragraphs + literal and linear translation of previously highlighted keywords, phrases and sentences
Highlighting keywords, phrases + sentences	Literal and linear translation of previously highlighted keywords, phrases and sentences
Highlighting keywords, phrases + sentences	Paraphrasing translation of previously highlighted keywords, phrases and sentences
Highlighting keywords and phrases in ST + schematic content overview	Literal and linear translation of previously highlighted keywords and phrases + filling-in or elaboration of schematic content overview
	Paraphrasing translation of previously highlighted keywords and phrases + filling-in or elaboration of schematic content overview
	Literal and linear translation of previously highlighted keywords and phrases + literal and linear translation of a selection of ST phrases and sentences + filling-in or elaboration of schematic content overview
Highlighting key sentences in ST + schematic content overview	Literal and linear translation of previously highlighted sentences + filling-in or elaboration of schematic content overview
Highlighting keywords, phrases + sentences + schematic content overview	Paraphrasing translation of previously highlighted keywords, phrases and sentences + filling-in or elaboration of schematic content overview
	Literal and linear translation of previously highlighted keywords, phrases and sentences + literal and linear translation of a selection of ST phrases and sentences + filling-in or elaboration of schematic content overview

The same is true for the four students who started their writing phase by translating several ST paragraphs “in full”. Two of them had not displayed any content weighing behaviour, such as highlighting, during the pre-writing phase. This implies that they may initially have deliberately postponed semantic reduction to a later stage (i.e., to the end revision). When they altered their approach during the writing phase, one of these students adopted a hybrid approach, switching between literal translation and paraphrasing for specific phrases and sentences, whereas the other student opted resolutely to paraphrase. In contrast, the other two students had exhibited semantic reduction behaviour prior to TT production, with one highlighting ST keywords, phrases, and sentences and the other drafting a TT outline. However, it can be argued that these strategies, which were initially considered to be planning strategies, could rather have served as comprehension strategies, since these students proceeded to translate the first six and seven paragraphs in full. To formulate the remaining TT paragraphs, both students adopted a literal patchwork strategy by translating ST phrases and sentences that they had previously highlighted or by simply integrating what they had already written in the content overview, respectively.

As stated in Section 4.1, 11 of the students had displayed an outline planning strategy, but only nine of them later used the TT outline in their writing phase. The two students who did not use the TT outline they had previously produced displayed different behaviour. One of them started translating the ST in full; the other had drawn up an extremely concise TT content overview, characterized by unclear cohesion between prominent content features and a mixture of source-language and target-language words. Moreover, entire paragraphs were not integrated in this outline. The quality of the outline (or the lack of it) may have prompted the student to re-initiate the ST content selection from scratch. Indeed, he proceeded by copy-pasting the ST in MSWord and started to transfer specific ST phrases and sentences into the target language, using a default literal translation strategy and deferring further content reduction by way of deletions in the end revision phase.

The seven students who did not exhibit any ST-content weighing behaviour in the planning phase also varied considerably in their drafting strategies. As mentioned above, two of these students started translating the first ST paragraphs in full, subsequently adopting either a literal translation strategy or a combined literal–paraphrasing approach. Among the remaining five students, three produced the TT using the paraphrasing strategy and verbalizing parts of the ST in their own words. They integrated content units from various ST sentences into highly idiomatic propositions in the target language, continually refining the TT formulation to ensure cohesion and coherence between sentences and paragraphs. In contrast, the other two students focused on specific ST phrases and translated them literally, with minimal attention being given to cohesion between propositions.

4.3 Summarization taking place during revision

The post-writing phase, which starts immediately after the writing phase and finishes when the participant ends the Inputlog logging session, is characterized by self-revision of the TT (Englund Dimitrova 2005). However, self-revision is not limited to this phase; it can also be performed during the writing phase of the TT produced so far (i.e., online revision) (Dragsted and Carl 2013). In fact, the examples illustrating the literal patchwork strategy and the paraphrasing strategy (Figures 5 and 6) have shown that summarization may also take place during online revision, primarily in the form of deletion. In the case of this study, it is therefore important to consider online revision as entailing ST semantic reduction and/or TT word-count reduction.

As shown in Table 5, 18 out of 26 student translators indeed already engaged in substantial revision during TT production. Of these 18 students, 16 also did end revision, whereas the remaining two did not (with the subsequent absence of a post-writing phase). In contrast, eight students exhibited extensive revision behaviour exclusively in the post-writing stage of the translation process. The revisions captured in the logging data suggest that half of the students (13 out of the 26) had a double revision focus when they were revising their TT during the writing phase or the post-revision phase, namely, both on summarizing (mainly aimed at TT reduction) and on possible ST interpretation and TT formulation considerations (e.g., spelling mistakes).

Table 5:

Frequency of online and end revision strategies.

Among the 18 students who engaged in substantial revision during the writing phase, 10 engaged in semantic reduction by deleting a significant number of TT propositions. Seven of them combined this online revision focus on deletion with dealing with matters related to ST interpretation and TT formulation. No clear preference was observed in the TT production and online revision strategies that were adopted: five of those 10 students who focused on reducing the TT word count by way of deletion in their writing phase had employed the literal patchwork strategy (exclusively or in combination with the “full” translation strategy or the strategy of filling in the TT outline); the other five students had used the paraphrasing strategy.

Sixteen of the 18 students who engaged in online revision also conducted one or more rounds of end revision. Among them, 12 students also engaged in summarization in the post-writing phase. In fact, most of them engaged in multiple rounds of end revision, each round focused exclusively on a specific issue. Figure 8 shows an example of a translation process with multiple rounds of end revision. In Figure 8, the post-writing phase starts at about 1h40, as is observable in the drop of the dotted green line (representing cursor movements); this means that the student moved from the end of the TT to the beginning. It is also observable in the slight decrease in the solid green line, which suggests the deletion of TT characters. An interpretation of the solid and dotted green lines indicates that multiple revision rounds can be observed after 1h40 – the first and third of which were concentrated on shortening the TT to adhere to the prescribed word count. Conversely, the second loop seemed to prioritize the idiomatic expression of the TT, albeit with fewer revisions having been conducted compared to those in the initial revision loop.

Figure 8:

Process graph of summarizing translation process with several end revision rounds.

In contrast, four students had already focused on deletion during online revision and concentrated instead on revising the ST interpretation and TT formulation matters during the post-writing phase. Two of them were literal patchwork translators, the other two paraphrased consistently during the TT production.

A similar finding was observed for the eight students who engaged exclusively in end revision: three focused on both TT reduction and revision, whereas three primarily deleted TT items while continuously monitoring the TT word count. Interestingly, those three students who focused exclusively on TT reduction in the post-writing phase had all employed the literal patchwork translation strategy (exclusively or together with the “full” translation strategy or the strategy of filling in the TT outline). No clear preference in TT production and revision strategy combinations could be established for the other three, who did not focus on summarization in their end revision. The same is true for the remaining two students, who focused exclusively on resolving ST interpretation and TT formulation problems in their end revision.