Disruption in Translator-Client Matching : Paid Crowdsourcing Platforms vs Human Project Managers

The paid crowdsourcing business model has brought a disruptive change to the translator procurement environment, offering clients algorithm-based automated systems as an alternative to conventional human-mediated project management services. This article analyses the conceptual/epistemological differences between the two from the viewpoint of knowledge management and considers the implications for future development of the industry.


Introduction: Disruption explained
In his seminal book The Innovator's Dilemma (1997), Clayton M. Christensen maintained that the kind of businesses that have been disrupted by newcomers are ironically 'wellmanaged' ones, which listen to their customers and aim to produce high-quality products and services.Christensen argues that market disruptors, on the other hand, use technologies to offer a lower-performance product (at least to begin with).By offering cheaper and more convenient products to a mass market, and gradually improving the quality of the product, disruptors eventually overtake existing wellmanaged companies.Christensen's case studies concerned computer disk drives, construction excavators and retailing processes, but the notion of disruption is now predominantly linked to digital online products, hence the concept 'digital disruption' (often held to be industry gospel) was born.
Translation crowdsourcing fits well the definition of digital disruption: it has been enabled by digital internet technologies; its service is simple and convenient to use; it is cheaper than conventional translation services (or sometimes free) and the quality of the end products is perceived to be lower (Flanagan, 2016: 160-162).The features seem to be perfect to cater for demands of emerging markets such as translations of shorter, non-specialised texts in e-commerce and social media.
Because the original concept of 'crowdsourcing' is closely linked to the notion of amateurs in general (Howe, 2006), many studies of crowdsourced translation have focused on amateur-oriented initiatives (e.g. Mcdonough Dolmaya, 2012;O'Hagan, 2009).However, since around 2008 language service providers (LSPs) started to use crowdsourcing platforms for their for-profit translation procurement activities (Garcia, 2015).Some scholarly research has already analysed its disruptive influence on conventional translation businesses (Flanagan, 2016;Garcia, 2015;2017, Jiménez-Crespo, 2017a, 2017b;Moorkens, 2017).Observation of crowdsourcing platforms, however, is not straightforward because of the difficulty of classification, partly due to the continuous evolution of crowdsourcing models, whether paid or non-paid (Jiménez-Crespo, 2017a: 11-12).Different classification systems have been proposed, which include Jiménez-Crespo's (2017a: 30-32) five-category model (participants' profiles, method of initiation, workflow, type of translation and type of initiative), Flanagan's (2016: 151-153) three-category model (provision/non-provision of payment, profile of crowd, initiator of the call for participation) or Garcia's (2015: 20, 26-27) continuum model, which positions crowdsourcing as an identifiable middle ground translation procurement method between machine translation and conventional LSP service.
However, as Garcia (2015: 27) admits, the boundaries between such classification groups are not rigid.
Although this process is understudied, it should be regarded as an important step in translation production as it serves as part of translation quality assurance (see section 3).This article analyses the use of platforms in this process in comparison with the project management method used in conventional LSPs.By doing that, I argue that the platform models overlook interpersonal knowledge in favour of metadata accumulation.
Despite the difficulty of defining due to the ever-evolving state of the industry, "paid translation crowdsourcing" is defined here as below, following Jiménez-Crespo's (2017a: 14) eight-point definition: An activity in which translation is outsourced by a for-profit company to the pool of registered translators on the company's platform, and for each job selected members of them, in a form of open call with a clear deadline for a payment.The company makes profits from the outsourcing activity and the outsourcing and delivering activities are carried out online on the Internet.This definition, however, does not guarantee that the companies discussed in this article engage solely with the activities defined here.Companies diversify their businesses in their own unique ways, which may make their business portfolios go beyond this definition (such as machine translation or data supply, as will be seen in section 4).Those activities outside this definition are, however, not the direct concerns of this article.

Paid translation crowdsourcing: A new business model in the platform economy
The conventional translation production process where an LSP receives an order from a client and commissions it to a freelance translator can be called a 'pipeline' business.It requires step-by-step arrangement for creating value, with the translator at one end and the client at the other (Parker, Van Alstyne, & Sangeet, 2016: 6).In this process, the project manager is the gatekeeper of the production process, who assesses the scale and complexity of each project and decides which translator to use from the LSP's pool of translators.Project managers' work is normally time-consuming and labour-intensive, and their performance often relies on their personal knowledge and experience gained in their work.
With paid translation crowdsourcing systems, this translator-client matching process takes place automatically on a digital platform without involvement of human project managers.Obviously different companies have different processes, but the following is a description of a typical process (using information available in Gengo, n.d. and Wisgo, 2017).A client uploads the source text on the company's website, chooses the level of difficulty (e.g.Standard/Business/Ultra) and language combination and receives an automatic quote.Once the client pays the fee and places the order, the system sends an email notification to translators who are eligible to take this job, depending on their language combination and translator status.Translator status is received by The system automatically matches the order to eligible translators (who are claimed to number in the thousands in many companies' marketing materials, such as Gengo, n.d.) depending on their statuses and the first translator who accepts the job by clicking the accept button is commissioned with the job.
In this system, the use of a "platform" is the key for maximising its effectiveness.In the most general sense, platforms can be defined as digital infrastructures that bring together different actors (Srnicek, 2017: 43).They let external producers and consumers engage in value-creating interactions (Parker et al., 2016: 5) and acquisition and mergers take place frequently (e.g.Slator.com'sM&A and Funding page list those ("M & A Funding," 2018)).This industry landscape suggests that, though the end-products of these services are cheaper than conventional ones, the model requires a high level of initial investment.
One of the purposes of the paid crowdsourcing model is to replace the conventional project managers' work with a translator-client matching algorithm on the platform.The use of a digital platform brings strong advantages to this replacement for the following reasons.First, translator-client matching is an immaterial and informational activity.Platforms are good at storing a large amount of data, thus good at knowledge-based work (Srnicek, 2017: 38).Second, platforms can capitalise on network effect, namely, the number of the users on the platform can have a positive effect on the value created for them (Parker et al., 2016: 17).The bigger the pool of translators, the more attractive the platform becomes for clients, and vice versa.Once this feedback loop is set in motion, the network can grow rapidly with minimum cost, which creates a disruptive effect on the existing industry (Parker et al., 2016: 65).And third, and most critically in the current discussion, platforms possess the privileged access to the metadata of the activities that take place during the interactions on the platform (Srnicek, 2017: 44).As clients request work, confirm their orders, receive translations and send their feedback, all the work-related metadata is recorded on the platform, including the translation itself (in the form of bitexts if the platform provides a translation memory system), the speed of the translator's work, the location of the translator, the time of the day (or night) the translator works, and the quality of the translation in the form of feedback from the clients.Garcia (2017: 66-67) points out that this data-gathering capability of platforms skews the power relations in favour of the owner of the data, i.e., the platform owner.
Once translators supply metadata about their professional profiles and productivity to

89
the platform and get used to the online environment (such as using the supplied CAT tool), their performance can be assessed solely using the metadata.Translators are then pushed towards using the platform as that is the only way of obtaining reputation (Garcia, 2017: 67-68).The key issue here is that translators are assessed automatically based on the data gathered, with a strong focus on productivity (How fast is the translation provided?How cheap is it?).The platform is designed so that translators feel strongly pulled to the system while the power shifts to the owner of the platform.
As digital economy scholar Nick Srnicek (2017: 46-47) claims, one of the general characteristics of platform businesses is that "[w]hile often presenting themselves as empty spaces for others to interact on, they in fact embody a politics".Scrutiny of the methods for using such metadata for translator-client matching by individual companies is beyond the scope of this study, but it would be reasonable to hypothesise that translation crowdsourcing platforms would be keen to utilise their power as owners of such metadata to continue the expansion of their operations.

Human project managers and their tacit knowledge
In the conventional "pipeline" business model, translator-client matching is carried out by human project managers.Olohan and Davitti (2015) provide a detailed description of how this is done in small-to-medium-sized LSPs, where project managers place importance on building trust between their translators and themselves.This is achieved through various methods, which include, according to Olohan and Davitti's workplace observation: making the translators feel they are in control by allowing them some freedom to suggest deadlines; taking care to use the translator's preferred mode of communication (e.g.phone or email); making sure their emails to the translators do not contain typing errors; and providing positive feedback explicitly but conveying clients' negative feedback in a toned-down language.These methods require strong interpersonal skills.Project managers ensure the implementation of these subtle but dynamic trust-building mechanisms because they are aware their relationships with their translators play an important part in the LSPs' quality assurance (Olohan & Davitti, 2015: 8).Experienced project managers know that using a reliable, highly-skilled translator ensures the delivery of a good translation product and that tactful interpersonal techniques are a necessary effort to secure such talented translators even if it is time-consuming.The importance of project managers' interpersonal skills and actions is also highlighted by translators, i.e., they affect translators' motivation and work attitude, and eventually quality of their translation (Rodríguez-Castro, 2013;Sakamoto, 2017).
Project managers build their knowledge of individual translators (their linguistic abilities, writing styles, professional behaviours, etc.) and related contextual information the translators' professional profiles, evaluation results of the quality of translation as well as project managers' experiences with them, both positive and negative (Sakamoto & Foedisch, 2017).However, a lot of the information is difficult to verbalise and will remain within the minds of project managers.This is known as "tacit knowledge" in the discipline of Information and Knowledge Management.
Tacit knowledge, as opposed to explicit knowledge, is "knowledge that is usually unverbalized and not explicitly taught" (Wagner & Sternberg, 1985: 437).According to Collins (1990: 4), there are two ways to learn knowledge: the "algorithmic model" and the "enculturational model".Within this framework, paid crowdsourcing translation platforms learn the art of translator-client matching in the form of explicit knowledge through the algorithmic model using metadata gathered on the platform, whereas translation project managers in pipeline-model LSPs learn the art (at least to some extent) in the form of tacit knowledge through enculturation.Collins (1990: 24) further maintains that enculturation is "the only way to master an expertise which is deeply laden with tacit knowledge because it is only through common practice with others that the rules that cannot be written down can come to be understood".It would be reasonable to argue that the intricate interpersonal actions of project managers as described above are possible because they possess the necessary tacit knowledge, which was gained through their professional experience.
4 Conclusions and future outlook I have analysed the differences of translator-client matching methods between the paid crowdsourcing business model and the conventional LSP model highlighting the underlying knowledge management systems.The quality of the product produced by paid crowdsourcing systems is yet to be evaluated, but project managers suspect the quality is low (Sakamoto, Rodríguez de Céspedes, Berthaud, & Evans 2017: 10).They believe that without correct judgement of human actors (i.e., project managers) it is impossible to make the best decision on translator-client matching.They are particularly concerned about the fact that the translator who presses the button the quickest after the job is placed on the platform secures the job on the paid crowdsourcing system (Sakamoto et al., 2017: 10).Anecdotal evidence suggests that this emphasis on speed in the 'first-come-first-served' principle of translator procurement places translators in a dilemma that careful consideration of a job offer (even by several seconds) risks the job being taken by competitors (Wisgo 2017).In the pipeline style project management, in contrast, human project managers can serve the role of advocates of the translators, who can protect them from this kind of pressure, and as the gatekeeper of the production process, who can control the quality of translation with their knowledge about both translation and translators.
These virtues of project management may not be a serious concern for paid crowdsourcing companies, though, because lower performance is what innovative platform services are supposed to offer (at least at the beginning) according to the tenet of digital disruption (Christensen, 1997).And project managers feel strong frustration and irritation about this very point.Platform entrepreneurs tend to believe  (Kenney & Zysman, 2016: 67).In project managers' eyes, paid crowdsourcing platform developers are attempting to establish a new logic of value creation in translator-client matching by keenly pursuing efficiency, productivity and scale, entering the market "unabashedly aim[ing] at serving not translators, but clients" (Garcia, 2015: 24).This challenges the project managers' belief in careful management of translators, which we saw in section 3. The two opposing approaches to translator-client matching are clashing head-on in their conceptual and epistemological principles.
The question is then, "Does it matter?"If we believe that the disruptive business model is meant to serve the lower-end of the industry, as Christensen (1997) originally argued, new entrants may not pose problems to existing LSPs: the two models would be able to cater for different sections of the market (as often called "premium" and "bulk" markets by industry insiders; see, amongst many blog posts, e.g.Hendzel, "Why translators are …") and live happily next to each other, or at least the damage to traditional translation businesses need not be significant.Flanagan (2016) reports this is what many professional translators believe.However, the rise of digital technology that thrives in the new order of the platform economy gives an unclear picture as to whether such segregation of the market would be realistic.At this stage, looking at the platform economy in general, "it is not clear whether these digital platforms are simply introducing digital intermediaries or actually increasing the extent of gig or contract work" (Kenney & Zysman, 2016: 63).
It is also worth noting that the paid crowdsourcing translation companies, which have been the disruptors in the industry, are now beginning to be disrupted too.For instance, the Pennsylvania-based paid crowdsourcing translation company u-Translate went bankrupt in February 2018 (Diño, 2018).uTranslate attributed their business failure to free machine translation services which can provide translations in the same time as their platform could match a client to a translator, drawing their customers to the "sheer, overwhelming, undeniable convenience" machine translation can provide, even at the cost of translation quality.In reaction to this environment, paid crowdsourcing translation companies are diversifying their business models.One Hour Translation has connected with a machine translation provider (One Hour Translation, 2018) and Conyac has started a voice data supply business (AnydooR, 2018).Even the disruptors in translation need to continue evolving their business models to defend themselves from yet other innovative technologies.
Translation is very much in flux for both new and conventional actors.This article focused on just one aspect of the current changes in the industry, i.e., translator-client matching, but it is important to recognise that the way translation is produced as well as translation itself will keep transforming through a combination of technologies.One such case is the simultaneous use of translation memory with a machine translation post-editing function on a crowdsourcing platform via a mobile app (Jiménez-Crespo, 2016).In this environment, the practice of translator-client matching may receive further changes made by other technologies.It is of our interest to continue observing after taking a competence test at the time of registration.

(
in what environment they work in what way) through the experience of working together with the translators, and the knowledge accumulated in this process functions as operational guidelines for them.Part of this knowledge becomes documented and shared amongst project managers.It is a common practice that an LSP builds an inhouse database, which contains information about their registered translators such as Akiko Sakamoto Disruption in Translator-Client Matching: Paid Crowdsourcing Revista Tradumàtica 2018, Núm.16 Platforms vs Human Project Managers 90 . Though there is now a vast variety of platform business models across different industries, wellknown examples are services such as Facebook, Uber and Airbnb (for detailed categorisations of platform businesses, see Srnicek, 2017: 50-88).In the translation environment, platforms are the mediators which enable translation clients and freelance translators to interact with each other so that the clients can receive translations from the translators.Currently such businesses include new venture capital entrants (e.g.Gengo since 2008, One Hour Translation since 2008, Conyac since 2009), new business divisions of existing LSPs (e.g.Lionbridge's onDemand, SDL's Language Cloud),