In search of a fair MTPE pricing model: LSPs’ reflections and the implications for translators

ABSTRACT

Machine translation post-editing (MTPE) has become an important part of the commercial translation business. As such, MTPE pricing models are under-studied yet vital elements in determining the professional value of translators as workers. This focus-group study investigates the discourse of eight LSP representatives relating to their MTPE pricing practice and, using the notion of psychological contract from organisational psychology, analyses what kind of external messages, including untold meanings and cues, translators may receive from the MTPE pricing practice. We then discuss how such messages may shape translators’ individual beliefs relating to contractual terms in MTPE projects, as well as their implications for MTPE operations in general. Three significant points are identified regarding; (1) How LSPs use different pricing models for different stakeholders to manage translators’ motivation for MTPE work; (2) How the notion of ‘speed’ incorporated in the MTPE pricing methods has a potential to determine translators’ work motivation and recruitment; and (3) How the power structure of technology ownership may influence future MTPE pricing practice. The outcomes of this exploratory study will contribute to the development of much-needed industry-wide conversation about this controversial topic.

KEYWORDS:

• Language service providers
• psychological contract
• machine translation post-editing (MTPE)
• MTPE pricing models
• translators
• focus group

1. Introduction

Low rates of pay for translation, in general, is a major source of worker dissatisfaction amongst translators.¹ The reasons for low pay may inherently come from the public’s perception about the work of translation, embodied in the conduit metaphor or the comparison with secretarial work (Lambert & Walker, 2022, citing Gouadec, 2007 and Katan, 2011). Other scholars argue that translation is affective labour, inherently prone to exploitation (Koskinen, 2020). Translators’ dissatisfaction is particularly prominent in the MTPE sector of translation services where a low-cost policy is an important sales strategy (Moorkens, 2017; Pielmeier & O’Mara, 2020). LSPs lament they cannot find enough skilled translators who are willing to take on MTPE work, yet at the same time, MTPE rates are an important determinant to motivate translators to take on this kind of work (Nunziatini, 2019). This complex dynamic is one of the key reasons why it is vital to investigate the practical MTPE pricing models, as we have done in this study.

There has been no industry-wide consensus about the best pricing method for MTPE; LSPs are left to devise their own pricing models (Nunziatini & Marg, 2020), which is why they have been finding MTPE pricing extremely difficult (Bayan Marketing Team, 2021). Furthermore, and perhaps due to its personal nature, the topic of money tends to be avoided in discussions amongst practitioners, and industry discourse lacks transparency on this topic (Lambert & Walker, 2022; Larsonneur, 2018). This is not helped by the fact that translation associations are discouraged from discussing translation and MTPE pricing issues as it might be regarded as a price-fixing practice, breaching anti-trust laws. In translation studies too, despite Gambier's (2014) call for an ‘economic turn’ in the discipline, the topic of ‘money’ in MTPE has been understudied, with a small number of exceptions (do Carmo, 2020; Vieira, 2020; also Lambert & Walker, 2022 for translation in general). Considering that the issue of payment is of paramount importance when seeking fair implementation of this relatively new practice in the translation industry, this study tackles the thorny topic of money in MTPE by first examining MTPE pricing models currently used in the industry. It then examines the discourse of LSP representatives about their MTPE practice to discover what external messages LSPs are sending to translators and how those messages may shape translators’ beliefs and attitudes about MTPE.

2. Theoretical frameworks

This study adopts two theoretical frameworks: one at the macro level, the other at the micro level.

2.1. Practice theory

At the macro (epistemological) level, this study engages with the practice theory of translation. Olohan (2021) introduced practice theory to translation studies, mainly through reference to the work of the sociologist Theodore Schatzki (2001). He defines practice as ‘embodied, materially mediated arrays of human activity centrally organized around shared practical understanding’ (Schatzki, 2001, p. 2). Practice theory prioritises practices to systems or norms as the target of observation when conceptualising and analysing the social world. It allows us to observe practitioners’ emotions and motivations not from individuals’ perspectives (as much research in translation studies has done) but as something that is constituted by a practice (Olohan, 2021, pp. 82–83). We consider LSPs’ MTPE pricing models as a practice that enables us to understand psychologically-laden, value-inducing aspects of MTPE operations in the industry, or what Schatzki terms the ‘teleo-affective structure’ of MTPE practice, in terms of their ‘ends, projects, tasks, purposes, beliefs, emotions and moods’ (Schatzki, 1996, p. 86). Examining a teleo-affective structure has strong practical implications, particularly where novices are socialised into a practice that is upheld by established practitioners (Nicolini, 2012, p. 166).

2.2. Psychological contract

On the micro (methodological) level, we draw on the concept of psychological contract from organisational psychology. We will explain what the psychological contract is further on, but the main rationale for adopting this concept is to avoid adopting a simplistic attitude to the topic of remuneration as something straightforward just because financial terms are stated numerically in the work contract. One of the common narratives within the high end of the translator market, namely, amongst translators well-established and specialised enough to be working mostly with direct clients, is that, as a freelance worker, the translator has the autonomy and freedom to choose the kind of work they want to take on by scrutinising the contract and specifications of the work project and negotiating financial terms with the LSP (Adams & Morris, 2014). However, this is not always possible for translators, particularly for professionally less established ones (Vieira, 2020). Instead of looking at the issue of (low and precarious) income as personal matters of individual translators, this study approaches it as a structural matter pertaining to the current translation industry by examining the psychological aspects of negotiations between different actors and agencies. The concept of psychological contract enables us to identify and observe relevant factors (both tangible and invisible) operating in negotiations between the agencies that may otherwise be missed, when for instance merely observing visible interpersonal interactions in workplaces or documentary evidence (such as work contracts).

This study particularly explores what factors may influence the way translators’ psychological contracts are formed in relation to MTPE work. This will be done by examining what LSPs say about MTPE pricing models. We believe that paying adequate attention to relevant factors involved in establishing pricing models will lead us to renewed and constructive considerations on the perceived value of MTPE work and, eventually, translators’ work motivation.

Now, what is the ‘psychological contract’? The term represents a concept derived from organisational psychology, which is defined as ‘individual beliefs, shaped by the organization, regarding terms of an exchange agreement between individuals and their organization’ (Rousseau, 1995, p. 9). It refers to ‘the perception of an exchange agreement between oneself and another party’ (Rousseau, 1998, p. 665, emphasis added), as opposed to the actual legal exchange of an agreement. In organisational psychology, the concept is typically used to study psychological states and processes involved in employer-employee relationships. The present study adopts this concept in the typical LSP-freelance translator relationship, as 75% of translators are estimated to work as freelancers worldwide (Pielmeier & O’Mara, 2020).

In this study, psychological contract denotes the translators’ psychological processing of the contractual terms of their MTPE work. It is important to note that the beliefs are about the ‘agreement’ the translator perceives themselves to have made with the LSP, thus these beliefs also have the power of self-fulfilling prophecies, eventually shaping the future development of an industry as a whole (Rousseau, 1995, p. 9).

According to Rousseau (1995), the founder of the theory of psychological contract, two sets of factors operate in forming the psychological contract of workers: external and internal. In an MTPE setting, we can assume that the external factors include external messages and social cues projected by the LSP or social setting. The internal factors include the translator’s internal interpretations and predispositions (p. 34).

An example of an external factor would be an advertisement slogan on an LSP’s corporate website about their MTPE services. An example of a cue emitted through the social setting would include a translator’s blog post describing their experience with MTPE work. External factors are not limited to overt corporate statements or messages. They can consist of direct observations of how colleagues are treated by the organisation (although this tends not to happen to freelance translators as they normally work from home). Other examples of external messages include expressions related to organisational policies (including manuals and handbooks), and references made by the general public connected to the history or reputation of the organisation and to events that happened in the organisation (Rousseau, 1995, pp. 36–40). Through these messages and cues, translators will form a belief as to how they are expected to perform in translation assignments, not only in terms of what is agreed in the written contract, but also in terms of how they should behave and how much effort they will expend.

In this study, we consider the LSPs’ MTPE pricing models and/or pricing-related discourse circulating in the translation community to be one of the vital external factors in creating translators’ psychological contracts. When messages or cues regarding MTPE pricing are perceived and interpreted (or ‘encoded’) by the translator, how that encoded information is used (or ‘decoded’) depends on the characteristics of the translator (Rousseau, 1995, p. 34), i.e., on an internal factor of psychological contract. We position this study within the exploratory bracket as a starting point for the investigation of translators’ psychological contract, so we focus on the external factors of psychological contracts, particularly examining how the external factors are created by the LSPs in their MTPE pricing practice.

3. MTPE pricing models

A review of the most relevant literature indicated that three pricing models are currently in operation in the industry.²

1. Hourly rate model

Translators are paid for the time it took them to complete the MTPE project. Translators will often have their own hourly rates, or LSPs may set a specific hourly rate for an individual project. The hours spent on post-editing can be recorded in two ways: self-report by the translator, or by recording using a programme such as a CAT tool.

2)	Per-word rate model (at a discount of translation rate)

This model was most frequently stated in the literature and documents we surveyed. For translation-from-scratch work (whereby no MT is used), freelance translators normally receive payment at their own translation rate, or ‘per-word rate’ (meaning one word in the source text costs x). In this MTPE pricing model, their per-word rate becomes discounted, typically at 25–40 per cent.

The precursor of this pricing model comes from translation work with translation memories (TMs), where the discount rate is decided based on the full and fuzzy matches provided by TMs. Although this pricing method has not been popular amongst translators, it has long been a standard practice (Zetzsche, 2019).³ Translators are paid at a discount rate of their normal translation per-word rate or a new project-specific rate which also tends to be lower than the translator’s normal rate. The discount can be applied on the source text word count or on the target text word count, although the latter is rare.

Discount rates can be decided in two ways. One is referred to as an ‘ex-ante model’, whereby the rate is decided without an evaluation of the MT output (Plaza-Lara, 2020, p. 170). The other method involves a small pilot project. For example, Scansani and Mhedhbi’s (2020) detailed report of their company’s discount rate involves two translators in the calculation of each project: the first evaluating the MT quality, and the second measuring the time taken for post-editing. However, Scansani and Mhedhbi (2020) state that this kind of effort had never been documented before to their knowledge and is rare in the industry.

3)	Post analysis (edit distance) pricing model

In this model, after the post-editing is completed, the LSP compares the text the MT engine originally produced (or a raw MT output) and its post-edited version. The amount of the text edited by the post-editor as against the number of the words in the raw MT output is called the TER (translation edit rate) or ‘edit distance’ (Snover et al., 2006). This can be represented by the number of words or a percentage of the raw MT output. The translator is paid for the amount of post-editing, i.e., the higher the edit distance is, the higher remuneration they receive. The edit distance can be measured in a CAT tool using its own algorithm, or the LSP may use their own.

Both academics and translators have criticised this method as being unfair for translators, as edit distance is normally not proportionate to actual post-editing effort (do Carmo, 2020; Hall, 2012) and Vieira and Alonso (2018) warn LSPs not to rely solely on this method.

In addition, we conducted an informal survey on the internet to collect translators’ personal opinions about MTPE pricing models which are publicly available online. Following Wilkinson and Thelwall’s (2011) method, the keyword command ‘site:’ was used, followed by the appropriate website domain name on the public part of the Google search engine through Facebook, Youtube, Flickr, WordPress, and Twitter. The keywords ‘MTPE’ and ‘machine translation’ were used to identify possibly productive sites, then more particular mentions of these issues were identified by searching the terms ‘pricing’ and ‘rates’. The top 10 ‘hits’ on each of the searches were surveyed manually to identify who was ‘owning’ the conversation and what they say about MTPE pricing. What we found in these searches is that MTPE pricing is rarely mentioned in public online spaces, being recognised as a very difficult and under-researched issue (Bayan Marketing Team, 2021; Cresceri, 2022). In addition, translators’ personal blog posts are extremely difficult to find compared to the more accessible corporate blog posts of technology companies, entities offering MTPE training and other interested parties such as translator associations. This is perhaps because there has been a shift in focus amongst the available blogs relating to translation, away from the opinions of individual translators and over to more promotional approaches by corporate LSPs and entities who use SEO (Search Engine Optimization) algorithms to seek high-rankings when Google is used as a search engine.⁴ Pricing is not specifically mentioned as an element in any of these, as the authors tend to be promoting their own views on technologies, training or talks at MT-related conferences, meaning the discussion is institution-led in some way by a company, training entity or association. The shift is perhaps also partly related to the implementation of the EU’s General Data Protection Regulation (GDPR) in 2018, when individuals in the language services sector have become aware of the need to protect data and their own professional reputation by using a certain amount of discretion in public communications on social media.⁵ The shift in the focus of publicly available blog posts (supposedly a democratic space where individuals can exercise freedom of speech) to corporate sites poses a methodological challenge in research, but it also has more serious ramifications for the shaping of industry discourse. We will come back to this point in our discussion later.

4. The study

4.1. Data collection

We held a focus group with representatives from eight LSPs to collect their opinions about MTPE pricing models. We sought, and gained, ethics approval from Kansai University, Faculty of Foreign Language Studies Ethics Committee (Ref 22-20) before conducting the focus group. The participating representatives were chosen by the Association of Translation Companies (ATC), a UK translation industry association and the research collaborator of this study, from ATC member companies that are active providers of MTPE services. Participating LSPs were based in the UK and other European and Middle-Eastern regions, but most of these LSPs were multi-national companies, holding offices in more than one country. They were mostly small- to medium-sized LSPs. The participants were CEO, managing director or chief operational officer (COO) of the companies.

The focus group was conducted on an online conferencing platform Zoom on 1 September 2022. It lasted 90 min. The first author was the first moderator, and the second author the second moderator. Online focus groups have been increasingly used due to the convenience these offer for participants who are spread over a large geographical area (Archibald et al., 2019). Care was taken to replicate the benefits of face-to-face focus groups (Abrams et al., 2015). This included clearly presenting ground rules of the group at the beginning of the session; asking the participants to keep the camera on during the discussions; encouraging participants to spontaneously unmute themselves and come into the discussion without raising a virtual hand; allowing us to carefully observe participants’ gestures and facial expressions, making sure to give all participants opportunities to speak.

The focus group was prepared and conducted in the following order. A week prior to the focus group, we produced a short, anonymous, multiple-choice style online questionnaire which asked the participants which pricing models (out of those identified in literature review) were being used at their companies and in what way. For example, in a question about the usage of the edit distance model, the respondents were asked to choose all options that applied to them from (1) ‘edit distance calculated using a function in a CAT tool’; (2) ‘calculated using the company's own algorithm’; (3) ‘calculated using any other calculation method’; and (4) ‘we do not use this method’. Six LSPs out of the eight filled in the questionnaire. The questionnaire revealed that the most commonly used pricing model was the per-word model, used by all of the six companies. Four of them set the per-word rate for each translator, one for each project, and one sets a rate either for each translator or for each project. Five LSPs set a per-word rate on the source text and one on the target text. The hourly rate model was being used by three companies alongside a per-word method. Two of them used the number of hours reported by translators and one reported by a software programme. Finally, the edit distance model was being used by three companies: two alongside a per-word method, and one along with both per-word and hourly rate methods. One more participant said during the discussion that their company also used this method, but only when they were billed by some of their MLV (multi-language vendor) clients.

These results were used to design the focus group’s question schedule; we asked the participants to discuss pricing models in the order of per-word rate method, edit distance method, and then hourly rate method. The results also informed us about the variety in the implementation of the methods. The two main questions that shaped the overall focus group discussion were: ‘Why do you use that method?’ and ‘Why do you not use that method?’ In case the discussion stalls, some more specific questions were prepared, as well as some prompting questions such as ‘Are there any points to add?’

At the focus group, all the eight participants joined the online conference promptly. An initial brief introduction where all participants showed their faces on the screen was followed by a lively discussion. One participant chose to turn off the camera and stayed silent most of the time, but all other participants were engaged throughout the discussion. The first moderator ensured time was allowed within the discussion for all three pricing models to be explored, but the direction and the depth of the discussion were led by the participants’ knowledge and interests. At the end of the discussion, we offered participants a chance to send post-group comments through an online form in case they had felt they could not express their opinions fully in the session.

4.2. The analysis

The focus group discussion generated a text of 10,460 words and an additional online comment of 77 words. The data was anonymised and then analysed using a thematic analysis approach (Boyatzis, 1998) with the software NVivo. A theme in thematic analysis is ‘a pattern found in the information that at minimum describes and organizes the possible observations and at maximum interprets aspects of the phenomenon’ (Boyatzis, 1998, p. 4). We aimed to identify patterns of information about MTPE practice as described and organised by the participants (at the minimum level of analysis), and then to give our own interpretation of those aspects by framing them through the lens of psychological contract (at the maximum level of analysis). Thematic analysis offers three different ways to develop a thematic code: (a) theory driven, (b) prior data or prior research driven and (c) inductive or data driven (Boyatzis, 1998, p. 29). Considering the characteristics of the data, the research question, and the theoretical framework we are adopting (psychological contract), a hybrid of all these approaches (Boyatzis, 1998, pp. 51–53) was used in this study, as described below.

The analysis was conducted in five stages. In the first stage, both authors read the transcript together while listening to the audio recording to grasp and discuss the overall arguments and sentiments involved in the focus group conversation. In the second stage, because the focus group discussed three different pricing models, the transcript was coded at the three labels named by the pricing models (Per-word rate model, Edit distance model, Hourly rate model). This process identified a forth model being used in practice, i.e., Hybrid pricing model, which was added to the list of codes. This analysis stage was primarily data driven. In the third stage, the transcript was coded without a predesigned list of codes, similar to open coding in grounded theory method (Glaser & Strauss, 1967). Here, interesting expressions used by participants were used as ‘in-vivo’ codes and developed into relevant themes by explicating the meaning (Barbour, 2018, p. 127).⁶ This coding approach was employed so that any concepts and ideas that have not yet been conceptualised in literature can be captured in the analysis. In the fourth stage, the themes represented in the codes generated in the third stage were compared with each other to identify meaningful and articulate themes (Boyatzis, 1998, p. 52). Similar themes were combined to form more abstract themes. The third and fourth stages of analysis were data driven and created 15 themes.⁷ Then, in the fifth stage, our analysis shifted to be theory driven. As the aim of our study was to find out what messages LSPs may be sending to translators that may shape their psychological contract about MTPE work, we searched for themes that can be interpreted to be functioning as external messages to translators.

In order to ensure sufficient inter-rater agreement, the first author analysed the data in stages two to five, then the second author checked the analysis in order to avoid the concepts being ‘contaminated’ by the researcher’s ideological views (Armstrong et al., 1997). Any discrepancies in the interpretation of data were discussed between the two authors to arrive at the results presented here.

5. Results and discussion

The analysis led us to identify three aspects of LSPs’ MTPE pricing practice that we deem significant in shaping translators’ psychological contracts, as presented below.

5.1. Pricing models as a means to negotiate inter-actor relationships

After closely examining the themes used to explain the three distinct pricing models, a remarkable finding was that decisions on pricing are not only about contractual agreements between LSPs and translators, but also that pricing models serve as a complex, and sometimes experimental, relationship management tool to control translators’ psychological contracts.

What became evident initially was that, when asked why they took part in the study, all the participants said they wanted to know what was happening in the industry. They took part in this study ‘to get a little bit more information about how other people are using [MT]’ (P3) and ‘to see if what [they] have come up [with] is actually sort of industry wide’ (P8) because they ‘have plenty of doubts’ (P2). This confirms our initial understanding that there are no norms in MTPE pricing practice at this stage.

Out of the three pricing models which were identified in the literature review, the participants were quick to clarify that hourly rates are least useful, because it does not allow clients to know the final costs of translation before they commission the work. While this model favours translators, it is, for their clients, ‘too up-in-the-air, especially when translation can often be quite a minor part of the budget for the overall project’ (P7). It is evident from this statement that the clients are the ones driving the pricing decisions and that the translators’ working conditions are being determined by the clients’ needs through the pricing models chosen.

Participants were more eager to talk about the per-word rate model as they have more experience with it. They emphasised the benefit of this model that (again) the final cost can ‘look more transparent from the client’s point of view’ (P2). They also thought that the discount rate needs to be ‘as fair as possible for the amount of effort that’s going into it’ (P7) but this is difficult because it ‘depends [on] what the (MT) output is to start with’ (P7). MT output segments are not guaranteed to have the level of consistency that can be expected between the sentences in a TM, meaning that the translator has to expend ‘a heck of a lot more effort because they’re trying to unify something that has no unity existing already in the base reference material’ (P7). Pricing becomes complicated especially in the increasingly common situation where TM and MT are used simultaneously in a CAT tool environment, where corrections of TM suggestions and MT outputs need to be priced separately. Two participants said this hybrid method, with their current pricing methods, has proved to be more profitable than a TM-only set-up. This suggests that achieving satisfactory profit margins from MTPE requires a complex pricing design.

Calculating an appropriate per-word rate is a challenge for LSPs. LSPs that have in-house translators can ask them to do a pilot project to decide an appropriate per-word rate. This is, however, not possible for LSPs who only use freelance translators. Nevertheless, the per-word rate model seems to be, overall, the most accepted pricing model despite some unclear aspects such as measurement of translators’ work effort. In the absence of a perfect method, LSPs are, ‘for now, … sticking with the per word rates’ (P5).

As described in Section 4.1, the pre-focus group questionnaire had suggested that post analysis (edit distance) was less commonly used amongst the participant LSPs. However, the discussion revealed that their use of this model is highly complex and controversial. Out of the four participants who used this method (alongside a per-word method), three LSPs used the method only for some of their clients, but not for their translators. Two LSPs even did so at the risk of making financial losses for themselves. One participant mentioned they let a client charge with this model as a test case, to learn from the experience. Another participant said their LSP used it for project assessment purposes only, but never for billing purposes.

This pricing model has been a source of contention, as LSPs feel translators are not being justly rewarded, supporting academic claims (do Carmo, 2020; Vieira & Alonso, 2018). The edit distance measurement tools currently available to them ‘simply measure the number of keystrokes’ and ‘don’t account for how long a linguist has had to actually sit and think about that sentence’ (P7). As a result, it can be ‘tricky’ and ‘confusing’ for translators and ‘the good linguists (i.e., translators) simply don’t want to do it’ (P7). There is also a risk that some dishonest translators may ‘somehow over-edit to get the right numbers’ of edits for themselves (P4). Another source of contention is that the final price can only be determined after the project is finished.⁸ This forced one LSP, who does not own a tool to measure edit distance, to delay payment to their freelance translators as this LSP could determine the payment amount only after the LSP was billed by their clients, saying this is ‘not the best method for the vendors (i.e. translators)’ (P1). One participant expressed their wish that a method could be established that can ‘define [edit distance] at the beginning of the project rather than post-project’ (P5). In addition, some participants admitted they either do not own the technology to measure edit distance or their understanding of its functionality is inadequate. These statements indicate that technology ownership is also influencing LSPs’ pricing decisions (see further discussions about predictive pricing methods and technology ownership in Section 5.3).

5.2. Post-editor recruitment and the meaning of ‘speed’

Recruitment is a theme directly related to the notion of psychological contract of translators. The focus group discussion suggested that LSPs’ post-editor recruitment uses two approaches: (1) using existing translators; and (2) recruiting new freelance workers who exclusively take on MTPE work. The first method could be challenging because some of their translators still do not have the ‘mindset to do post-editing’ (P1). One participant, however, said it is now ‘an easier sell to get’ (P3) because improved MT quality enables translators to post-edit high volume translations within a short timeframe, making it possible for them to earn more money than when translating from scratch. This refutes an earlier belief amongst LSPs and translators that post-editing is poorly rewarded work, suited only to less qualified translators (Sakamoto, 2019).⁹

It is important to bear in mind that this profitability of MTPE is only a reality if translators work at high speed, and for many larger-scale LSPs this means introducing cloud translation platforms where post-editors’ working speed is constantly monitored by ‘a lot of algorithms in the back of [them]’ (P3). None of our participant LSPs had introduced such platforms yet, partly because of the high up-front costs of installing them. The underlying principle of such a system is that, whether the payment is based on the per-word rate or the post-project analysis, the faster they work, the more money they earn. Interestingly, our participants showed two contrasting opinions about this principle. One participant said it would be ‘scary’ for translators to be placed under constant surveillance, being treated as if they were ‘becoming machines’ (P1) to receive higher earnings. This harms the autonomy and wellbeing of the translators. Still, another participant recognised that, although translators ‘don’t like somebody monitoring them 24/7, if they get paid 25% more for doing that’, that works as ‘the carrot […] for the translators’ (P3). These contrasting opinions suggest that ‘speed’ has nuanced and divisive meanings amongst LSPs.

High-speed work can also have an impact on quality. If translators are ‘being paid for speed, [they] don’t research any more’ (P4) when, for example, the MT output comes up with an unknown term, resulting in a poor-quality translation. This would be acceptable where the client requires only ‘good enough’ translation, but the majority of the LSPs’ clients ‘are wanting MT but still want perfection’ (P7), meaning that an excessive focus on high speed may lead to client dissatisfaction.

These outcomes allow us to assume that, when translators do post-editing on an LSP platform, they will feel their professional value is being measured by the speed of their work. At the same time, they will be pressured to work fast to receive higher earnings. In this situation, the external messages translators receive would be that their worker autonomy and well-being are undervalued by the LSPs. In addition, the work arrangement would signal to translators that quality is not a priority, which may encourage the translator to compromise their work quality. This suggests that intense emphasis on speed of work could have strong implications for translators’ psychological contracts.

Speed can also be a difficult and risky notion to manage for LSPs. Let us think of an example where an LSP claims on their website that they provide high-standard MTPE services to customers,¹⁰ but they implement a pricing model for translators where the translators are rewarded for speed while sacrificing worker autonomy, well-being and work quality. This is a common phenomenon known as ‘the folly of rewarding A while hoping for B’ (Kerr, 1975; cited in Rousseau, 1995, p. 182), where the translators will receive a message that differs substantively from that given to the client. Rousseau (1995, pp. 182–183) states that it is normally human resources practices, such as payment terms, rather than client-facing policy, that send the strongest message to workers, shaping their day-to-day behaviours. This misalignment is likely to harm their motivation to take on post-editing jobs, or to do these well if they do take them on, producing serious ramifications for LSP recruitment operations.

5.3. Technology ownership and the role of big technology companies

The previous section touched upon the issue of surveillance of translators which is enabled by a platform-based work environment. It is particularly noteworthy that this platform-based workflow with an intensive expectation for high speed is something the participants felt is imposed by ‘big boys’ (P3), i.e., large LSPs that own large resources and advanced technologies.¹¹ Our participants, who were small- to middle-sized LSPs, felt they were unable to develop such a platform-based system because ‘It costs quite a lot of money, in the millions, to be able to get that working’ (P3).

In addition, those ‘big boys’ may not necessarily be language specialists. Large technology companies, which were originally non-linguist companies, have been advancing in language technology areas such as machine translation: some participants mentioned Amazon as an example. What is worrying for LSPs is that large non-linguist technology companies may bring ‘processes and software that you use in other (industries) into our industry’ (P3). In order to gain competitiveness under such circumstances, a number of mergers and acquisitions have recently been taking place in the translation industry.¹² In general, our participants found this development ‘scary’ (P3).

In contrast, one participant expressed a cautious welcome to the big boys’ presence, saying smaller LSPs may benefit from their advanced technologies in the future. One of the reasons why establishing an effective pricing model is difficult, after all, is that each post-editing project is unique. It is thus impossible to devise a ‘one size fits all’ pricing model that can ‘come to the right fee that covers the effort’ of the translator / post-editor (P7). Therefore, instead of trying to come up with a perfect one size fits all pricing model, large technology companies may ‘evolve and allow [smaller LSPs] one day to choose the pricing model that we will want to use on a specific workflow’ and have it ‘automatically be integrated in the workflow’ (P2). More specifically, they suggested that the ‘big boys’ may develop a big-data-driven algorithm-based pricing tool which can automatically predict an optimal pricing method for each project. If such a function were to be integrated into a TMS (translation management system), LSPs would be freed from the complicated process of deciding a pricing method for each project.

We put aside the discussion of whether this kind of TMS system is technologically possible at this point in time, as our aim is not to examine technological possibilities. Instead, we raise two important points that should be critically considered if such an automated pricing tool becomes reality.

First, the development of such a system should involve smaller LSPs, like our participants, with their own experiences and insights into individual translators’ skills and professional motivations. The resultant tool will otherwise very possibly provide a skewed model towards larger LSPs which are willing to promote a ‘the-faster-the-better’ work ethos. No technological artefacts are inherently neutral and the design of a technological tool reflects the developer’s political and ideological stance and, most importantly, shapes the social and economic consequences for all users of the tool (Winner, 1999). Olohan (2011) theorised how translators interact with technologies inspired by Pickering’s (1993) notion of ‘mangle of practice’, using a case involving CAT. The technological artefact gains a certain position within the work environment through interactions involving resistances and accommodations (which Pickering called ‘the dance of agency’). One deciding factor for resistances and accommodations in such interactions is whether the human agency has a say in the development of the technology. For example, in a study of MT adoption in two different workplaces, Cadwell et al. (2018) observed that a new technology is more easily accepted when the users are involved in the development of the technology.

Our initial online documentary survey revealed that online discourse surrounding MTPE is predominantly corporate-driven, particularly by larger LSPs. Online discourse is also increasingly shaped by technology giants using their SEO algorithms (Google Search Engine is the most prominent example). It is our concern, in this landscape, that a (currently hypothetical) automatic MTPE pricing tool may reduce the range of payment options available to smaller LSPs and the translators / post editors working for them, while excluding them from a broader discussion as their voices are not sufficiently visible in industry discourse. Consequently, an uncritical adoption of such an automatic pricing tool by LSPs may send a message to their translators that their historical LSP-translator relationship is no longer valued.

Second, although our participants emphasised that translators’ work effort should be fairly reflected with regard to MTPE pricing models, it is important to point out that any prediction of translation effort is very difficult, both theoretically and methodologically. Against the backdrop of the ‘predictive turn’ in translation studies (Schaeffer et al., 2019), much cognitive research has been conducted with the aim to devising a method of predicting post-editing effort before it is actually carried out, by using knowledge about the human translation process and features of MT outputs. There is nonetheless still ‘a huge gap between existing translation-oriented theories and models and their operationalization with methods from cognitive science’ (Schaeffer et al., 2019). Examples of such gaps include discrepancies in methodological principles between automatic and human quality assessments, and limitations in ecological validity in research experiments (Schaeffer et al., 2019). Ogawa (2021, p. 17) also warns that the attempt of measuring translation ‘difficulty’ using the concept of ‘effort’ may be theoretically flawed and has a risk of oversimplifying the discourse about translation and its difficulty. How the translator feels about the difficulty of the task and how they distribute their effort depends on the translator’s experience (Hvelplund, 2016) as well as their skills and work circumstances. So, if we aim to devise a fair MTPE pricing model based on effort to be expended, person-related elements need to be incorporated (Ogawa, 2021; Sun, 2015). This, of course, is theoretically and methodologically challenging.

While a theoretical and methodological consensus about translation and post-editing effort is lacking, it would be extremely optimistic, or even dangerous, to expect that big technology companies will be able to develop a ‘fair’ pricing tool. Such an expectation would be an epitome of solutionism (Morozov, 2013), i.e., the belief that complicated social problems can be solved simply by using modern technology. In fact, two participants showed a sign of solutionism in our focus group discussion, wishing that large LSPs and technological companies might give smaller LSPs a solution to the complicated problem of MTPE pricing. If this attitude becomes prevalent in the LSP community, this may send a signal to their translators that their professional value is now defined by a technology-oriented, but theoretically unjustifiable, understanding of translation.

6. Conclusion

We have investigated the discourse of eight LSP representatives about their MTPE pricing practice in order to examine what kind of messages they may be sending to translators. Drawing on the concept of psychological contract (Rousseau, 1995), we considered such messages are important because they influence translators’ psychological states and the processes involved in their interpretation of their LSP-translator relationship, their expected work performance, and eventually their work motivation and future career development. Three significant points were identified: (1) MTPE pricing models can be (and are being) used as a complex relationship management tool by LSPs, although not always to their advantage or by choice; (2) The divisive notion of ‘speed’ incorporated in the MTPE pricing methods has a potential to determine translators’ work motivation and recruitment; and (3) The industry discourse about the value of MTPE services is increasingly being led by large technology-driven LSPs, and the creation of future MTPE pricing models (and the utilisation of technology in the designing process) can be a major influence in the upcoming power dynamic between tech-adept and tech-limited LSPs. These outcomes enabled us to assess the possible risks of an uncritical ‘solutionist’ adoption (Morozov, 2013) of data-driven technologies for translators’ psychological contract.

MTPE is a complex translation production system and this study focused on one aspect of it, i.e., MTPE pricing models. In the spirit of practice theory (Olohan, 2021; Schatzki, 2001), it aimed to understand translators’ (hypothetical) emotions and motivations as influenced by the pricing practice of LSPs. More specifically, we examined which kind of external messages LSPs are sending to their translators through the practice of MTPE pricing and how they ‘may’ influence translators’ psychological contract, or how translators may ‘encode’ the messages (Rousseau, 1995, p. 34). Further studies will need to investigate how such external messages actually affect the way the translators behave (i.e., the ‘decoded’ aspect of psychological contracts: Rousseau, 1995, p. 34). Such studies will require observations of internal factors such as characteristics of translators or their cognitive processes. We thus hope this study can provide a useful starting point for further studies as well as a much-needed industry-wide conversation about the complex and understudied topic of MTPE pricing.

Acknowledgements

We would like to thank the LSP representatives who willingly took part in this study in their busy schedules. We would also like to thank the ATC (Association of Translation Companies) for their help with the arrangement of the focus group.

Disclosure statement

Akiko Sakamoto declares no conflict of interest. Sarah Bawa Mason has a part-time, paid consultancy role as Commercial Collaborations Lead at the Association of Translation Companies (ATC).

Additional information

Funding

This work was supported by JSPS [grant number JP22K20040].

Notes on contributors

Akiko Sakamoto

Akiko Sakamoto is Professor in Translation Studies at Kansai University, Japan. Before moving to Kansai in 2022, she taught and researched translation at the University of Portsmouth. Situated in translation studies and drawing from science and technology studies, she investigates what impacts technologies have on translators’ work flow, job satisfaction, career motivation and social status. Her recent work include a handbook chapter ‘Translation and Technology’ in The Cambridge Handbook of Translation (2022), Cambridge University Press, and a book chapter (with Masaru Yamada) ‘Managing Clients’ Expectations for MTPE Services Through a Metalanguage of Translation Specifications: MPPQN Method’ in Metalanguages for Dissecting Translation Processes (2022), Routledge.

Sarah Bawa Mason

Sarah Bawa-Mason is a Senior Lecturer in Translation Studies at the University of Portsmouth and also works in consultancy mode as Commercial Collaborations Lead for the Association of Translation Companies. She was Chair of the Institute of Translation and Interpreting (ITI), UK, from 2016 to 2019. She continues to run the ITI Research Network and she is a regular invited speaker at international translation association and industry events. Sarah has taught Specialised Translation at the University of Portsmouth, Bristol University and London Metropolitan University. She publishes on the topics of translator training, professional aspects of translation and the potential impact of new technologies on the profession.

Notes

1 Following do Carmo and Moorkens (2021) we regard post-editing as a kind of translation. Therefore, we use the term ‘translator’ even in the context of MTPE, unless it is appropriate to use the word ‘post-editor’, or both, in a specific context.

2 Literature covering this area is limited in translation studies, but Ginovart Cid (2020) and Plaza-Lara (2020) provide some information about MTPE pricing. An online seminar provided by Ingrada (2021) is another useful information source.

3 One translator cited in Zetzsche (2019, p. 193) calls this pricing method ‘wage theft’. For deliberation on the reason for the acceptance of this method by the industry, see Sakamoto and Yamada (2020),

4 For example, a representative listing from Speakt (https://speakt.com/translation-blogs/) of 50 leading translation blogs includes only a handful of blogs by individual translators and it is sometimes exceptionally difficult to work out who the ‘writers’ are behind contributions to these pages. A search of the named blogs by individual translators provided no discussion of MTPE pricing.

5 See, for example, https://www.iti.org.uk/discover/events-calendar/gdpr-and-the-uk-2021.html

6 Examples include ‘the carrot (for translators)’ which developed into a code ‘incentives’ and ‘big fish’ and ‘big boys’ which became a code ‘big techs’, a sub-theme of ‘Stakeholders’.

7 The themes (and the number of segments coded) are: Stakeholders (30), Not sure (or lack of knowledge; 14), Speed (10), Fairness (8), Incentive (or lack of-, 8), Quality of post editing (7), Recruitment (6), Effort (5), Measuring tools (5), Quality of MT (5), Transparency (5), Surveillance (4), Relationship (3), Wellbeing (3), Solutionist (2). However, the numbers of segments are not directly relevant to our interpretation of data, because the final analysis stage was theory driven.

8 This seems contradicting from the viewpoint of the hourly rate model discussed above; clients do not like the hourly rate model because the final cost cannot be determined until after the project is finished. This is probably because with the hourly rate model, it is difficult for a third party to control translators’ speed of work during the project, while edit distance can be numerically measured from the final text.

9 Note that Moorkens (2022), however, warns that this earning potential may last only for a short term depending on how the industry practice develops.

10 An informal internet search shows that some LSPs emphasise improving quality of MTPE services (e.g., https://www.stepes.com/machine-translation-post-editing/) while others are more cautious about claiming high quality (e.g., https://www.oneword.de/en/what-is-mtpe/). [Accessed on 22/06/2023]

11 Three LSPs that are placed within the top 100 LSPs (CSA, 2021) were mentioned by the participants as examples.

12 See, for example, Slator.com’s Merger and Acquisition reports at https://slator.com/ma-and-funding/.