This study explores how the process of translating relates to other types of writing processes by comparing pause lengths preceding syntactic units (words, phrases and clauses) in two types of writing task, a monolingual text production and a translation. It also discusses the grounds for interpreting pause length as a refl ection of the cognitive demands of the writing process. The data was collected from 18 professional translators using the Translog keystroke logging software (Jakobsen/Schou 1999). Each subject wrote two texts: an expository text in Finnish and a translation from English into Finnish (Immonen 2006: 316-319). Firstly, phrase boundary pauses were categorised according to type, function and length of phrase. All three features correlate with pause length. On average, predicate phrases are preceded by short pauses, adpositional phrases by long pauses, and pauses preceding noun phrases grow with the length of the phrase. These fi ndings suggest that the processing of the predicate begins before written production of the clause is started, whereas noun phrases and adpositional phrases are processed during writing. Secondly, pauses preceding clauses were categorised with respect to clause type. In monolingual text production, pauses preceding subordinate clauses are on average shorter than those leading to main clauses. In translation, pauses preceding subordinate and main clauses are almost the same length. It seems therefore that, in translation, the main clause and subordinate clause are processed separately despite the fact that the subordinate clause functions as a syntactic unit within the main clause.

This paper is an empirical study on pause patterns in fluent translation and monolingual text production. By comparing pauses recorded from both processes, it discovered two temporal features: Firstly, the mean length of pause at textual category boundaries grew the higher the category was in the syntactic hierarchy. Secondly, the length of pause at clause level or lower were on average longer in translation than in monolingual text production, whereas pauses above clause level tended to be shorter in translation. Besides the differences in pause duration, translation also affected the use of total production time. Translation requires on average a longer revision and monitoring phase while the drafting phase is completed more quickly. Both writing tasks used approximately the same proportion of time for the orientation phase.

This research focuses on the type and proportion of the revisions made by translators during target text production. Eighteen professional translators used the keyboard logging software Translog while carrying out a translation task on a computer. As Translog registers and displays all keyboard activity in relation to time, all the revisions in the log files can also be identified and counted. The revisions were categorised as revisions of typing errors, revisions of literal translation and other revisions. Typing error revisions account for 51.5%, literal translation revisions for 20.5% and other revisions for 28.0%. If typing error revisions are ignored, literal translation revisions account for 42.3%. Revisions of literal translation were observed at all linguistic levels and in all translators’ log files, irrespective of the quality of their final translations. These results suggest that literal translation constitutes an integral element of the translation process and can perhaps be considered as a strategy to expand the translator’s working memory. The results also give support to the Monitor Model of translation, which maintains that literal rendering is the default strategy of target text production.

No abstract