Live subtitling through respeaking makes live events accessible to diverse audiences, such as d/Deaf and hard-of-hearing people. Respeakers utilise automatic speech recognition (ASR) software, which they train to recognise their voices, thus improving accuracy, converting live speech into subtitles in the same or another language.
Respeakers use both textual and translation skills during their task, such as enunciation, psycho-cognitive and metalinguistic skills, and diamesic skills, that is, the ability to report speech in writing (Eugeni, 2008). However, there is still a gap in our knowledge on how we should teach respeaking and what kind of strategies students need in order to cope with the cognitive challenges of the process (Arumí-Ribas and Romero-Fresco, 2008). In this article, I discuss different respeaking skills and how they should be approached in teaching the method.
In order to identify the respeaking strategies, I chose to analyse the intralingual live subtitles from an accessibility conference entitled Les Assises Régionales de l’Accessibilité. I studied the data using Eugeni and Gambier’s (2023) taxonomy for the analysis of diamesic translation. The taxonomy comprises four set of strategies: litteratim, verbatim, sensatim, and signatim. The litteratim strategy concerns phonetical, phonological and orthographical tactics, rendering the source text phonemes (sounds) into graphemes (letters) in the target text. The verbatim strategy deals with the source text’s lexical and morphological aspects, with a view to delivering word-for-word subtitles. The sensatim strategy focuses on the source text’s syntactic, semantic and pragmatic tactics and is concerned with delivering meaning. Finally, the signatim strategy aims to deliver the source text’s paraverbal and non-verbal elements, such as prosody, gestures and postures of the speaker.
Eugeni’s and Gambier’s taxonomy proved comprehensive in evaluating the conference. In the conference, there was a prevalence of verbatim and sensatim tactics. This was particularly evident in the respeakers’ behaviour during their task and in the exit strategies that they used, that is, the techniques applied when a speaker is particularly challenging. The tactics used by a professional respeaker during the task are a key element of the design of a formal live-subtitling course. In the next section, I will discuss the key skills that should be taught to respeakers.
Teaching Respeaking Skills
The diamesic translation analysis showed that there is a need for sensatim-focused respeaking training. as respeakers mainly found themselves using tactics such as reformulation, compression and expansion to deliver the speaker’s message into the subtitles. In the light of this, I will delve into the production strategies that respeakers employ to convey their message.
Enunciation skills are paramount for quality target text delivery. Respeaking students must become familiar with ASR software, and must create domain-specific vocabularies and macros by identifying technical and uncommon terms present in the source text. The analysis showed that many domain-specific terms utilised during the conference were not present in the ASR software language model. Identifying them required prior preparation and training. Such terms included proper names (e.g., Stéphane Bourdon, Yvette Molina), names of other bodies or associations (UNIACCES, SENSOCOM), place names (Langoëlan, Morbihan) and domain-related terms (audioprothésistes ‘audiologists’, presbytie ‘farsightedness’).
For psycho-cognitive or multitasking skills respeakers must master the ability to speak and listen at the same time, reformulate, edit and correct their respoken output. Moreover, while listening simultaneously, they must remember complete sentences while lagging behind the speaker and deal with extralinguistic elements, such as slides or videos.
Diamesic skills comprise metalinguistic and compression skills. In cases of high source text speech rate, the respeaker needs to deploy synthetic skills to ensure readability. To make the target text accessible and comprehensible, the respeaker must develop skills for breaking the lines, adding punctuation, and implementing non-verbal elements, such as slides and videos. The diamesic translation analysis showed extensive use of punctuation (820 elements), following French syntactic rules.
Genre skills require students’ training in genre analysis. By analysing several texts of the same genre in detail, it becomes possible to determine a register, terminology and structures that are repeated and can even be anticipated.
Accessibility skills are critical in a respeaking training programme, as every speech needs to be adapted to the target audience’s needs and expectations. Accessibility also comprises the notion of inclusion. Respeakers at Le Messageur, for example, are committed to inclusive language where gender representation is concerned by retrieving the French male and female forms (ceux et celles) to address the audience.
Once production strategies are mastered, it is necessary to focus on strategies for tackling source text difficulties. There are various situations that may cause trouble to respeakers during their task, such as fast, incomprehensible, low-volume or impromptu, speeches or those including much play on words (LTA, 2018).
When trying to subtitle a fast speaker, a respeaker cannot produce an accurate rendition at the same speed because the professional is delayed by 1) the task of adding punctuation; 2) the strain on working memory capacity; and 3) the ASR’s recognition ability, which requires dictation to be made at an even pace and with frequent pauses. One exit strategy that can be applied is the Gordian knot strategy, which means subtitling every other sentence while maintaining coherence and providing understandable subtitles (LTA, 2018). In one instance during the conference, the respeaker kept the overall coherence and cohesion of the text even though some crucial sentences went missing in the process, such as other organisations’ names and proper names. For fast speakers, the Garwood strategy suggests leaving out unclear sentences and resuming from the first intelligible one (LTA, 2018). In the case of an overlapping exchange between two speakers, the respeaker decided to apply the Garwood strategy by omitting some unnecessary propositions, such as repetitions (e.g., oui oui oui ‘yes yes yes’) or filler words (e.g., absolument ‘absolutely’), preserving clear sentences that made sense in the target text.
If the respeaker does not understand one or more sentences, as for incomprehensible speech, they can opt for the generalisation strategy (LTA, 2018). Using this strategy, they may provide the correct information via a reformulation or by saying something general or logical when the context and topic are known. When the respeaker is unsure of the content, he can opt for the aforementioned Garwood strategy for incomprehensible speech. When a respeaker tries to caption a low-volume speech, he can resort to the Gordian knot strategy, hearing and repeating a sentence as fast as possible, compressing it, skipping the second sentence and resuming from the third to maintain cohesion and coherence with the first sentence (LTA, 2018).
An impromptu speech is characterised by features of orality. In the conference, these features took the form of the subject’s anaphoric retrievals (such as elle … elle ‘she … she’), hesitations and grammatical mistakes. For wordplays, respeakers typically stop live subtitling and type them in uppercase letters, exclamation points or ellipses (LTA, 2018). However, the analysis showed that these elements are generally omitted.
In this article, I have emphasized the importance of textual and linguistic strategies in respeaking, focusing on the intralingual aspect of this practice. I have also showed that it is important to have a holistic approach to the teaching and training provided for respeaking. Such an approach encompasses understanding, production and exit strategies, which are essential for respeakers if they are to deliver a quality target text in an inclusive communicational context.
Martina A. Bruno is a PhD candidate at the University of Bologna with a research proposal on museum accessibility through easy-to-read language and machine translation. She is also a professional respeaker and a trained conference interpreter.
- Arumí-Ribas, M. & P. Romero-Fresco (2008). A Practical Proposal for the Training of Respeakers1. Journal of Specialised Translation. URL: https://www.jostrans.org/issue10/art_arumi.pdf
- Eugeni, C. (2008).Le sous-titrage en direct : aspects theoriques, professionels et didactiques. EUM.
- Eugeni, C. & Y. Gambier (2023). La traduction intralinguistique : les défis de la diamésie. Timișoara: Editura Politehnica.
- Live Text Access (LTA) (2018). IO 1 report: Skills and competences. URL: http://velotype.com/LTA/IO1_LTA_Report_Skills_Competences_SDI.pdf