Public Studies (TPR-DB)

A keystroke dataset consisting of English-Dutch news translations. One professional and original translation is used as metric reference.

None

A test of error highlighting in post-editing to see if it decreases cognitive effort. Same data as DG21 but it includes error annotations by a professional EN>ES translation reviewer.

This study records interpreters' eye movements during reading the English transcripts of the source text, displayed in Translog-II.

This study records student translators' eyetracking and keylogging data in English-Chinese from-scratch translation.

This is the re-aligned version of STML18 based on the BOLT guidelines.

None

Compares students and professional translators as they revise human translations (English-German) that were manipulated to introduce errors of six different categories: orthography, grammar, meaning, omissions, consistency, and coherence. Two manipulated versions of each text were created (A and B) to ensure even error type distribution.

Eyetracking and keylogging data obtained from monolingual reading and copying tasks.

English-Japanese; translation dictation (D)

Eyetracking and keylogging data collected in from-scratch English-Spanish translation.

The ENJA15 study consistes of three different conditions: from-scratch translation, translation dictation, and post-editing. Pariticpants translated two texts (from the MultiLing set) in each of these conditions.

This study identifies the working styles of student translators in the process of self-revision, other-revision and post-editing (English-Chinese).

The IMBi18st study, re-aligned according to the BOLT guidelines 

Some participants translated without being able to see the target text they were producing as they typed it (they just saw an empty window), and other participants were able to see the target text they were producing as they typed. It is not currently clear how the researchers distinguished which sessions were which condition because they only used a single modality abbreviation: T.

An eyetracking and keystroke logging dataset obtained from from-scratch translation, post-editing of machine translation and monolingual editing tasks.

Eyetracking and keylogging data are available for from-scratch translation from English to German, whereas only eyetracking data are available for sight translation, monolingual copying, and reading aloud.

This study investigates how translators' mental processing resources are put to use during translation

This study was used to investigate English-Spanish neural machine translation (10 different NMT systems) of the MultiLing news texts with subsequent automatic word alignment using SimAlign's Match method.

Eyetracking and keystroke logging data obtained from from-scratch translation and post-editing tasks by professional translators.

A comparison of post-editing, translation, and sight translation with students working from English into Arabic.

This study was used to investigate English-Spanish neural machine translation (10 different NMT systems) of the MultiLing news texts with subsequent automatic word alignment using SimAlign's Argmax method.

Compares from-scratch translation, post-editing, and monolingual editing of news texts (English to Spanish)

Professional and student translators' gaze behavior while translating was analyzed in order to determine differences in reading behavior. Results were compared with simple reading gaze data (reading modality data is not part of this study and is not hosted on the TPR-DB).

Eyetracking and keylogging data are included in this English copying study. The source texts are from MultiLing.

This study, in conjunction with ATJA22, investigates the impact of translation briefs (full vs. light post-editing) and search conditions (termbase vs. internet search) on post-editing (PE) quality produced by participants with different levels of translation proficiency.

This study was used to investigate automatic translation word alignment of the BML12 study's original human translation, post-editing, and monolingual editing sessions using SimAlign's Intermax method.

It examines typing behavior in from-scratch translation and post-editing of machine translation.

This study compares a translation (T) or ‘no MT’ condition with post-editing (P) of NMT

This study, in conjunction with ATZH22, investigates the impact of translation briefs (full vs. light post-editing) and search conditions (termbase vs. internet search) on post-editing (PE) quality produced by participants with different levels of translation proficiency.

This study was used to investigate more consistent manual translation word alignment (human aligners followed strict alignment protocols) of the BML12 study's original human translation, post-editing, and monolingual editing sessions.

English-German; from-scratch translating with retrospective protocol (TA) and without (TB)

This study examines the impact of type of cohesive chain on cognitive effort in Portuguese-Chinese translation and post-editing tasks.

This study examines the differences between post-editing Google neural machine translation and from-scratch translation of English texts to Chinese

English <-> Brazilian Portuguese: L1 and L2 translations from/to English and Brazilian Portuguese.

This is the re-aligned version of STC17 based on the BOLT guidelines.

This study investigates the cognitive effort in the problem solving process guiding decision making in the translation of linguistic metaphors in a journalistic text by twelve professional translators from Chinese to Portuguese.

copying (C), paraphrasing (H), 4 participants, first 6 texts

This study was used to investigate automatic translation word alignment of the BML12 study's original human translation, post-editing, and monolingual editing sessions using SimAlign's Argmax method.

This study was used to investigate automatic translation word alignment of the BML12 study's original MT output (SMT generated by Google in 2012) using SimAlign's Match method.

Investigates parallel translation tasks selected from the College English Test Band-6 in order to determine what text features are the cause of difficulty.

Compares English-German from-scratch translation, bilingual post-editing, and monolingual editing of machine-translated texts.

This study tests the hypothesis that translation involves parallel rather than sequential processing of the source and target texts.

Spanish-English: post-editing of four news texts.

This paper investigates the effect of decontextualised source texts on behaviour by comparing post-editing of sequentially ordered sentences with shuffled sentences from two different texts.

Project redBird investigates correlations between human and machine translation output (for three target languages: Arabic, Spanish, and Japanese) in terms of translation entropy and source-text characteristics such as word class, figurative expressions, passive or active voice, and anaphora. See also ARMT19 and JAMT19.

This study was used to investigate automatic translation word alignment of the BML12 study's original human translation, post-editing, and monolingual editing sessions using SimAlign's Match method.

English-to-Chinese, translation & copying

This study investigated how orthographic and semantic similarity affect word production time in monolingual English text production tasks (copying, paraphrasing, and summarizing),

Project redBird investigates correlations between human and machine translation output (for three target languages: Arabic, Spanish, and Japanese) in terms of translation entropy and source-text characteristics such as word class, figurative expressions, passive or active voice, and anaphora. See also ARMT19 and ESMT19.

This study examines the differences between post-editing Google neural machine translation and from-scratch translation of English texts to Chinese

None

The IMBi18 study, re-aligned according to the BOLT guidelines 

Uses back-translations from two pivot langauges (texts were translated with NMT into Mandarin and Arabic and then post-edited by professional translators; then participants were asked to translate from Mandarin and Arabic back into English) to determine how pivot-language translations impacted translation accuracy. Automatic quality scores (such as COMET and BLEU) were compared to manual evaluations (Likert scores of human judgment of English back-translations' similarity to the original English source texts).

The study used data from a German-to-English translation experiment involving ten popular-science news ticker texts, each with a heavy (≥5 words) and a non-heavy (1–2 words) subject version. All translators were given one version of each of the texts, with the heavy and non-heavy stimuli being counterbalanced.

The data is used in Husam's PhD thesis, which examines whether the Need for Cognition (NFC) has an impact on cognitive effort and translation errors and how NFC interacts with text difficulty to influence cognitive effort and translation errors.

The English–Chinese translation experiment comprised two matched sessions – translating while thinking aloud and translating silently.

This study involves four translators and uses 50 English source texts. One professional and original translation is used as metric reference.

Eyetracking and keylogging data collected in English-Chinese from-scratch translation.

English-Estonian: 5 participants translating 3 different texts.

This study investigated the degree to which translators’ construction of a full mental representation of the source text prior to translation and their translation experience affect the literality of produced translations.

Google nueral machine translation output from English to Chinese annotated with error types by 16 translation students.

This study was used to investigate automatic translation word alignment of the BML12 study's original MT output (SMT generated by Google in 2012) using SimAlign's Intermax method.

Short German source texts were manipulated to have either short noun-phrase subjects or long noun-phrase subjects to see if translators preferred simpler translation solutions for longer, more informationally loaded noun phrases, all while studying the cognitive effort associated with translating each type of noun phrase and translators’ respective translation strategies.

Ten Dutch human translations of the multiLing set.

This dataset examines neural machine translation and human-machine parity in terms of evaluation and post-editing.

Unknown

This study was used to investigate automatic translation word alignment of the BML12 study's original MT output (SMT generated by Google in 2012) using SimAlign's Argmax method.

The data from this study was used in Annette C. Sjørup's doctoral thesis titled “Cognitive effort in metaphor translation: An eye-tracking and key-logging study”

A test of error highlighting in post-editing to see if it decreases cognitive effort.

None

This study examines the differences between post-editing Google neural machine translation and from-scratch translation of English texts to Chinese

Project redBird investigates correlations between human and machine translation output (for three target languages: Arabic, Spanish, and Japanese) in terms of translation entropy and source-text characteristics such as word class, figurative expressions, passive or active voice, and anaphora. See also ESMT19 and JAMT19.

In this dataset, the Google machine translation output into Spanish of the 6 English texts (the MultiLing set) was error-annotated by 8 professional Spanish translators. The annotations were based on MQM-derived error taxonomy.

Keylogging data collected during the production of news by two Spanish journalists.

None

This study compares different ways of presenting the texts on screen for post editing, namely with sentence and paragraph segmentation

unknown

Participants (students and professionals) translated (English to Norwegian) a text that included possessives that would force them to choose between reflexive and non-reflexive possessive pronouns in order to observe processing differences as they handled these different structures.

This study was used to investigate English-Spanish neural machine translation (10 different NMT systems) of the MultiLing news texts with subsequent automatic word alignment using SimAlign's Intermax method.

None

This study examines English-Chinese from-scratch translation and post-editing data from 21 first-year Master and MTI students.

An experiment designed to discover if more frequently occurring grammatical constructions are translated more quickly than less frequently occurring grammatical constructions.

This study compares eyetracking and keylogging data among post-editing of machine translation, simultaneous interpreting with source text, and from-scratch translation from English to Chinese.

Machine translation output using MultiLing source texts.