The freelance translation machine: Algorithmic culture and the invisible industry

Abstract

Much of the work performed by the global translation industry is handled by freelance labor. This segment of the industry has seen a radical structural transformation that has accompanied a radical transformation in the media environment that supports its work. The emergence of online freelance translation marketplaces has married the logics of standardization, automation, and protocol to casual labor, motivated by incremental profit and lubricated by entrepreneurialism. Customs and practices native to contemporary internet culture generate a freelance translation machine made of equal parts flesh and silicon that manages skilled labor algorithmically. In parallel with the specific case of freelance translation practices, this article develops and deploys a notion of algorithmic culture that accounts for the integration of human cognition in computational processes. Consequently, the possibility emerges that users instrumentalize algorithms even as algorithms instrumentalize users.

Keywords

Agency algorithmic culture entrepreneurialism freelance labor marketplace networks technology and society translation industry web

Contemporary translation practices sit at the interface between the digital and the analog precisely because they subject language to both the algorithmic logics of computer processing and the non-binary labor of human cognition. Even as machine translation is increasingly available and reliable in the form of no-charge, online utilities such as Google Translate, the work of human translators remains prominent in today’s translation landscape. This human labor, in the form of freelance translation, is delivered through database-driven online marketplaces such as ProZ.com that cultivate an image of automated, frictionless translation.

This article develops the notion of a freelance translation machine, an assemblage of circuits and flesh that transforms text from one language to another with a computer’s efficiency and the sensitivity of the human mind. Much attention has been devoted to the development of reliable, computer-powered translation that renders source text into another language via cryptographic manipulation.¹ However, contemporary freelance translation practices simultaneously embrace and reject the notion of language as an algorithmic, mechanically replicable configuration of rules, tools, and parts. Sites like ProZ.com act as digital matchmakers, using Web 2.0 technology to connect translation seekers to human labor that is contingent, geographically distributed, and lubricated by entrepreneurialism. Along the way, a fee is collected from the freelancers, injecting a new mediating layer into the translation economy.

Translation plays an increasingly important role in the circulation of culture, goods, and capital. The size of the translation industry is difficult to calculate due to amorphous activity boundaries (what separates translation from editing or proofreading?), ill-defined industrial sector boundaries (does the industry include work performed by people in other jobs or only by those identified as ‘translators’?), and a lack of reliable statistics: estimates range from US$3 billion to US$30 billion (Wooten, 2008). An average of recent estimates places the translation industry between US$7.3 billion and US$13.2 billion with an annual growth rate between 5 and 7.5 per cent (Boucau, 2005: 10, 13, 40). The three largest language service providers alone accounted for US$3.8 billion in revenues in 2009 (Romaine and Richardson, 2009). Freelancers are estimated to account for as much as 80 per cent of the industry (Boucau, 2005: 36). Despite the uncertainty associated with these figures, the translation industry can safely be characterized as large and the use of freelance labor within the industry as significant. This article explores how the freelance translation marketplace operates, how it reflects and illuminates other labor practices, and how it configures the relationship between humans and computers.

The freelance translation machine relies on the logic of the computer, the exemplary apparatus of globalization, to manage the manipulation of language. Under globalization, computers analyze, represent, and mediate the social, political, economic, and cultural exchanges of everyday life. They translate the smooth, continuous experience of analog life into discrete, discontinuous data that yield to the decision trees of computation. Their governing logic is that of algorithms, sets of ‘step by step instructions, to be carried out quite mechanically, so as to achieve some desired result’, which are often associated with computation, though they ‘are not confined to mathematics’ (Chabert et al., 1999: 1). But sometimes algorithms need help. Algorithmic culture reaches beyond the network and recruits human labor and cognitive power in a sort of extra-dimensional do loop. The effect is to bracket human work and make use of non-binary thought. The freelance translation machine executes such a maneuver, negotiating the encounter between the computational and the human in the service of capital. But by inviting humans to participate, the algorithm invites the possibility that it, too, will be used. Indeed, the algorithm’s successful unrolling depends on human translators’ ability to derive satisfactory levels of utility from their participation in a scheme that balances human labor and computational machinery against the backdrop of global political economy.

The extra-dimensional do loop

Critiques of algorithmic culture explore the human–machine interface with an eye to the entanglement of network architecture and the messy materiality of human culture. The category of algorithmic culture enables an understanding of both how computational logic pervades contemporary culture and how it shapes the possibilities of life itself. The two most prominent proponents of this idea are Alexander Galloway, who connects it to a Deleuzian notion of control, and Ted Striphas, who finds algorithms making determinations of cultural value. Here, I nuance and expand their critiques, arguing that algorithmic culture is not entirely algorithmic.

‘[T]o live today’, Galloway writes, ‘is to know how to use menus’ (Galloway, 2006: 17). Computer programs are not only applications, but also models for life itself. Their menus present predefined and strictly delimited actions: open, save, spellcheck. To live through this logic is to understand life as a series of moments where algorithms are selected from menus and executed. Galloway allegorizes the ideology of one system (software governed by algorithmic protocol) and imports it into another (life governed by algorithmic protocol). Applying this logic to a turn-based video game, Galloway explores the ramifications of imagining human characteristics such as race and nation as packages of standardized attributes switched on or off in varying combinations. Players structure their play around different racial and ethnic identities hardcoded into the gamespace. Galloway’s concern is that ‘the complexity, variation, and rich diversity of human life’ are erased (Galloway, 2006: 97). The game ‘transposes the many-layered quality of social life to an inflexible, reductive algorithm for “civilization”’ (Galloway, 2006: 98). The game thus presents itself as an allegory for a mode of cultural, social, and political control that Galloway (2004) theorizes as ‘protocological’. Galloway’s algorithmic culture is the cultural configuration where humanity’s round pegs are forced into the square holes of computer subroutines.

Where Galloway focuses on the ways that software serves as an allegory for a method of control, Striphas (2011) understands algorithmic culture as ‘the sorting, classifying, and hierarchizing of people, places, objects, and ideas using computational processes’, a binary approach to organizing a world inhabited by analog beings. For Striphas, algorithmic culture is opposed to its predecessor, elite culture. Whereas elite culture allows ‘a small group of well-trained, trusted authorities’ to determine what ought to be read, heard, seen, learned, and thought, algorithmic culture depends on ‘a statistical determination of what’s culturally relevant’ (Striphas, 2010). In this construction, services such as Amazon gather information about what users find important through more and less obvious collection instruments, aggregate the responses, and synthesize the data to pronounce on what to buy. These pronouncements are validated by the imputation that they represent an automated and dynamically updated measure of public opinion. Striphas’s critique is rooted in two related concerns: that new media technologies enlist users to provide the raw data used to shape tastes and that the processes by which these determinations are made are hidden behind intellectual property laws.

Both Galloway and Striphas’s critiques of algorithmic culture are machine-centered: they are largely concerned with the ways algorithms construct regimes of power and knowledge. They each show how algorithmic culture imposes binary logic on human thought, constrains its possibilities, and models its behaviors. These aspects of algorithmic culture are all part of ‘a war on surprise’, a widespread ‘endeavor to bring everything into the realm of governance and productivity’ (Aneesh, 2006: 23–24): human culture bends to computer specifications. A human-centered critique of algorithmic culture focuses on the ways that algorithms attempt to absorb human cognition without annihilating it. At stake is a more subtle exercise of algorithmic power, one less concerned with reshaping, reformatting, or restraining culture than with harnessing human thought, precisely because it does not conform to machine logic. Where Galloway’s algorithmic culture tells human thought, ‘Conform to my logics’, and Striphas’s says, ‘Tell me what you want so I can sell it to you’, mine says, ‘Go forth. Think… And then come back quickly with your reply’.

A do loop is a programming subroutine that executes a particular block of code until a specified condition is met. It is an algorithm’s algorithm. Contemporary freelance translation practice is an example of an extra-dimensional do loop inasmuch as it relegates part of the translation task to human minds. Source-language text is ejected from the binary network and transformed by humans into target-language text, which is then reimported into the system to complete the cycle. By subcontracting with humans, the freelance translation machine adds an analog dimension to binary computation: the human translator becomes a subroutine in globalization’s translation program.

Translation and globalization

The encounter between translation and computer networks under globalization generates a new arrangement of economic actors whose relationships are simultaneously interdependent, reinforcing, and antagonistic. If the human translator constitutes the extra-dimensional do loop, what can we make of the algorithm she completes? The freelance translation machine not only mixes human and computational power to move ideas across linguistic boundaries, it also introduces a new layer of mediation into the translation industry. Embodied in ProZ.com, this new mediating layer restructures the industry and recalibrates the flow of money, siphoning off a portion in the process. The result is a redefinition of the agency’s role, a sharpened focus on project management, and an increased demand of entrepreneurialism on translators to lubricate the machinery.

The freelance translation machine’s goal is the sort of invisible translation described by Lawrence Venuti (1995). Christine Mitchell (2010: 97) suggests that this elusive invisibility is bound up in ‘conceptions of the translation task which position the translator as medium’. In this sense, the translator – especially the non-literary translator – becomes one more interface between sender and receiver through which source text passes. Translators, like interfaces, are meant to be neither seen nor heard: ‘the more a dioptric device erases the traces of its own functioning (in actually delivering the thing represented beyond), the more it succeeds in its functional mandate’ (Galloway, 2008: 931).

But translators, like interfaces, are real, embodied components of social worlds. N. Katherine Hayles’s (1999) account of information’s inescapable embodiment offers a crucial point of comparison. Her clearest illustration is an anecdote from the planning of a cybernetics conference. A central figure calls for assistance: ‘“Take a letter, Miss Freed”, he says’ (Hayles, 1999: 82). But even if the scholar imagines that the letter is commanded and then appears, Miss Freed knows that ‘words never make things happen by themselves… For that, material and embodied processes must be used’ (Hayles, 1999: 83). Similarly, words never translate themselves. Even as the freelance translation machine presents the illusion of automation, there is always a man or woman² behind the curtain.

This shift in freelance translation comes against the backdrop of new consumer-grade machine translation software and experiments in crowdsourcing. Google Translate and Babelfish are the two most recognizable free, web-based automatic translation products. The Google model invites users to edit the processed text, harvesting suggested improvements to fine-tune the mechanism (Google, 2012). But these products remain famously unreliable, frequently failing round-trip translation tests (Search Engine People, 2011). A recent study of more advanced machine translation systems shows that, while programmers have developed programs that return increasingly accurate translations, ‘when style is included as a parameter of quality, then human translation is preferred’ (Fiederer and O’Brien, 2009: 69). To address this problem, companies like Facebook recruit users to produce translations, though accuracy and style haunt these translations, too (Hosaka, 2008). One possible solution is the use of crowdsourcing both to translate and to edit. Two computer scientists, Zaidan and Callison-Burch, experimented with Amazon’s Mechanical Turk, a microlabor platform where users perform microtasks for microwages (Scholz, 2009). They concluded that ‘it is possible to obtain high-quality translations from non-professional translators, and that the cost is an order of magnitude cheaper than professional translation’ (Zaidan and Callison-Burch, 2011: 1228). Each of these arrangements – Google Translate, Facebook, Mechanical Turk, ProZ.com – are essential cultural software for the globalized moment: automated interfaces for the organization of work that rely on human subroutines to complete tasks.³ ProZ.com’s difference lies in the expectation (and motivation) of some approximation of professional-level compensation for translators: even if the ProZ.com marketplace drives translators’ wages down, it does not lower them to Mechanical Turk levels. This expectation (and motivation) is paired with the expectation that those who enlist ProZ.com translators will receive professional-level work. So long as style stymies machine translation, algorithmic marketplaces like ProZ.com are the most reliable option under globalization.

Translation studies has proven adept at talking about globalization. As globalization is the heading we place above stories about ideas, goods, services, capital, and labor that dash across political and linguistic boundaries, it is a natural topic for a field dedicated to the transformation and transport of words. But the encounter between translation studies and globalization has so far been incomplete, owing at least in part to a blind spot in the field: a reluctance to engage with how the practices of translation fit into the networks of capital, labor arrangements, and communications technologies that have made it central to globalized culture. This blind spot could be explained by a fear of the ‘dirty money’ for which translators work (Hermans and Lambert, 1998: 116). The problem with translation studies has been translation, the use of a practice to designate the boundaries of a discipline (cf. Acland, 2003: 46).⁴

Michael Cronin’s Translation and Globalization (Cronin, 2003) shows a welcome willingness to address a split in translation studies between, on the one hand, work descending from literary studies, philology, philosophy, and linguistics focused on literary translation and, on the other hand, highly practical professional training that is preoccupied with increasing translator productivity. Translation studies boutiques have cropped up at universities around the world to teach and develop the technological tools that increase throughput and reduce turnaround times. For Cronin, it is the product of translation – the translated text – that organizes this strand of thought. His contribution critiques the shift from mass to flexible production of translations. But Cronin offers surprisingly little reflection on how the economic sector underlying these practices – the translation industry – has transformed and been transformed by shifts in translators’ working conditions. As Brian Mossop notes, this change, ‘while certainly enabled by information technology, is being driven by business pressures’ (Mossop, 2006: 790). Missing is the parallel story of the shift from mass labor to flexible labor. ‘The translator’, Mossop observes, ‘is still not at the center of [translation studies]’ (Mossop, 2000: 46).

Anthony Pym confronts the various economic interests surrounding translation. He argues (Pym, 1995) that translation is an option chosen by clients who find it less costly than learning a second language and that cooperation (the pursuit of mutual benefit for two or more parties) is a core goal of translation. Pym warns against using his cost-benefit analysis ‘to describe everything that happens or could happen in the field of translation’, suggesting instead that it illuminates ‘some of the ways that translation can contribute to [cooperation]’ (Pym, 1995: 602).

Meanwhile, Robinson (1997) responds to Pym’s cooperation thesis by noting that ‘filthy lucre is one of many tabooed subjects within the Western tradition of translation theory’ (Robinson, 1997: 11). Robinson argues for a translation theory that sidesteps the usual formalistic questions, all of which he believes are premised on the notion of some ideal method of translation that the theorist must seek out. Instead, he proposes a socioeconomic approach to translation studies that accounts for the ends to which translation is put, the forms of social stratification it engenders, and the conditions that lead to its production and consumption.

Elsewhere, Pym reflects on changes in the translation industry, particularly as its focus shifts to an ability to sell products wrapped in language across linguistic settings. Localization emerges (Pym, 2004) as a process of linguistic transfer adjacent to but distinct from translation, destabilizing the assumption that translation moves a unit of language from one single language to another single language. Localization is both the rendering of text from one language into another (or others) and the deployment of linguistic expertise to adapt dynamic data in meaningful ways to linguistically different consumers.

Marc Charron (2005) portrays localization as ‘translation in spite of itself’ (Charron, 2005: 16) and argues that localization demands a compression of time as well as space. For Charron, ‘localization distinguishes itself from translation because it moves more quickly than translation, because it presupposes a dizzying acceleration of translation’ (Charron, 2005: 17). Ultimately, Charron declares that localization is what allows language to catapult money to the grandest stage imaginable, and it is the temporal immediacy and the scope of the operation that characterizes translation under globalization.

But for most translation studies scholars, translation and globalization meet in the study of texts, tools, or methodologies. The field has been blinded by pioneer Gideon Toury’s 1995 assertion that ‘it is only reasonable that a study in translation activities… would start with… first and foremost the translated utterances themselves’ (Toury, 1995: 36). Even as translation scholars have theorized translation and its role in globalization processes, there has been a lack of reflection on how structural shifts affect translators, how the mobilizations of the terms ‘translation’ and ‘translator’ have changed, and how the network technologies that are intertwined with globalized translation have altered the ways language is commoditized.⁵ In short, what has been missing in translation studies is the translation industry.

Disintermediation and balance

The post-war translation industry had a relatively stable dramatis personae. Mossop inventories the ‘traditional business forms’, which included ‘the translation department in a government or corporation’ and ‘the freelance or small agency serving a local market’ (Mossop, 2006: 789). The dominant model today tilts the balance by relying increasingly on freelance labor. A decade ago, Janet Fraser and Michael Gold found that, since the 1980s, ‘[l]arge… companies… have been divesting themselves of in-house translators’ (Fraser and Gold, 2001: 683). In a later study, they surveyed British freelance translators and found that many were content with their departure from organizational employment and would not accept a full-time position (Gold and Fraser, 2002: 586). They reported coming to appreciate freelancing’s flexibility and the ability to determine one’s own working conditions (Gold and Fraser, 2002: 590). Read together, these two studies suggest that as organizations have reengineered translation off payroll, translators who shifted to portfolio work have adapted, sometimes discovering new rewards.

A similar change occurred among US translators in the same period. The American Translators Association periodically conducts a survey that is completed mostly by ATA members. In 2001 (American Translators Association, 2003), about 52% of respondents were full- or part-time independent contractors. In 2007, a member of the Association’s leadership wrote that freelancers ‘account[ed] for more than 70% of ATA’s members’ (Racette, 2007). None of this is to suggest that freelancers now perform all translation. Due to security, credentialing, volume, legislative, legal, and other constraints, some sectors of the translation industry continue to rely on staff translators with traditional contracts. However, these figures suggest that the rise in global demand for translation has been met by a rise in freelancers wherever possible and cost-effective.

This shift to freelancing did not correlate with any declining need for translation and resulted in a spike in demand for freelance translation. Fraser and Gold report that the British market was flooded with work. The survey data they collected from UK translators in 1999 suggested that freelancers enjoyed diversified client portfolios, the luxury of dropping clients who imposed unreasonable deadlines or paid late, an enviable degree of control over rates (Fraser and Gold, 2001: 690), and a mean income of more than £25,000 (about US$36,000 when Fraser and Gold’s work was published, about US$47,000 in 2011) (Fraser and Gold, 2001: 692).

But the story of translation’s embrace of freelance labor is not as simple as all that. The familiar plotlines of downsizing appear here, but there has also been a fundamental reconfiguration of the industry provoked by changes in communication technology. In simplified terms, the ‘traditional’ arrangement under which the translation industry operated resembled many other sectors where an intermediary connected workers to customers:

Translator \leftarrow \to Agency \leftarrow \to Client

This model applied to both freelance translators and those who were full-time employees of agencies. In either case, the agency provided translators with work and offered end clients some value-added service, usually project management functions or professional expertise that would be burdensome to manage in-house. The flow of cash and services was straightforward: the client paid the agency, which kept a cut and passed the remainder to the translator, who performed the work and handed it off to the agency, which then returned it to the client.

A key challenge faced by all three parties in this arrangement is disintermediation. Once a translator has performed work through an agency for a client, the agency is keenly interested in ensuring that the translator and client do not circumvent the agency. By specializing in the management of translation projects, the agency can convince clients that it is worth paying a premium to avoid the hassles of managing freelancers or maintaining a translation department. Similarly, translators find it advantageous to take pay cuts to maintain good relationships with agencies, which provide regular access to work. In effect, agencies thrive by preventing disintermediation while end clients and translators must balance the short-term benefits of dealing directly with one another against the long-term security provided by allowing agencies to manage supply and demand.

This solution to disintermediation reveals a specific relationship among translators, agencies, and end clients that functions under specific cultural, economic, and technological conditions. In an environment constrained by geography and limited by the number of linguistic combinations demanded by a relatively fixed set of clients, agencies were able to hold both end members of the translation chain at bay.

This balance has been upset over the last decade by the logics of networks and increasingly complex supply and distribution chains. The number of potential translation purchasers grew and diversified along with the number of economic actors interested in markets with different linguistic and cultural parameters. Pym (2006) constructs these trends as ‘an extension of Ricardan trade’ where specialized production leads parties to exploit comparative advantage (Pym, 2006: 747): ‘Products have to be moved from the specialized places in which they are produced; their information thus has to cross linguistic and cultural borders; documents have to be translated’ (Pym, 2006: 748). Simultaneously, globalization unrolled a set of information and communications technologies that allowed for the transmission of documents over great distances, enabling people to seek one another out and send messages freely and easily. But the Web’s effect on the translation industry has not been to disintermediate. It has instead introduced new mediating layers.

As the geographies of the translation industry grew beyond local markets and time-tested relationships, the need emerged for efficient methods of pairing far-flung clients and translators. Michael Cronin documented one early attempt to meet this challenge, a website called Aquarius that bills itself as the ‘longest standing online marketplace for translation and localization’ (aquarius.net, 2010). Cronin (2003) wrote that ‘the company itself, through the control of the subscription process…, acts as a gatekeeper with respect to individuals or as switch in the case of agencies or localization firms or client companies’ (Cronin, 2003: 47). Cronin’s interest in Aquarius arises in the context of a discussion of the differing levels of access to the internet, which determines who does and does not perform translation work. But Cronin’s analysis excludes the fundamental structural changes that online marketplaces like Aquarius impose. Unlike the three-party arrangement that characterized Mossop’s ‘traditional’ translation market, the networked translation industry includes a fourth member:

Translator \leftarrow \to Online Marketplace \leftarrow \to Agency \leftarrow \to Client

In this model, the online marketplace cuts into the chain. Now, clients come to agencies with work. Agencies evaluate the language pairs needed and the areas of specialization required. In the event that the agency does not have a translator on staff or a freelancer on file who matches the required profile and is available, the agency goes to the online marketplace. Freelancers in the marketplace submit bids; the agency selects the translator or translators with the appropriate skills and with whom the agency can agree on a rate. The work is performed, the agency is paid, and a portion is passed along to the translator.

But the online marketplace introduces not only another layer between translator and text. It also introduces a new transaction: to bid on the project, translators must typically first pay a membership fee to the marketplace.⁶ With an additional party and an additional transaction, the structural underpinning of the translation industry is reconfigured.

The key to the new equilibrium is that the middle layers do not compete directly since their revenues derive from different sources and are collected at different moments. Instead, the agency and the marketplace are mutually dependent. The marketplace needs agencies to supply work to the translators (the marketplace’s primary customers). Agencies need the marketplace to provide access to a large, linguistically diverse labor pool such that any given language pair will be represented. Moreover, agencies rely upon the online marketplace to grant them access to a geographically dispersed labor pool so that they can take advantage of downward pressures that a globalized workforce applies to wages. Meanwhile, clients are increasingly dependent on the agencies: with an exponentially larger array of translators available to meet their markedly more complex linguistic needs, agencies provide project management functions and sift through the piles of virtual resumes. Finally, the freelance translators are reliant upon the marketplace to provide access to work. In effect, each member in the linear relationship is reliant upon the adjacent players. But the balance is maintained because, although clients and freelancers might still benefit from direct relationships in certain circumstances in the short term, disintermediation threatens the end members as much as the inner players. In the reconfigured translation industry, a new structural arrangement also yields new cultural meanings.

The poetry of ProZ.com

Aquarius boasts so precisely of being the ‘longest standing online marketplace for translation’ because it cannot claim to be the largest or most significant. Fifteen years in, the field of online translation marketplaces is dominated by ProZ.com, a Syracuse, New York-based startup that marries the logics of internet marketplaces and Web 2.0 social networking (Beninatto, 2006). ProZ.com claims that more than 5000 jobs flow through the site each month (ProZ.com, 2011). By comparison, Aquarius claimed in 2010 that 10,415 jobs were posted at its site in the previous four years (Aquarius.net, 2010). The structure of the globalized translation industry is given voice in the language of the ProZ.com paratext. By closely reading the texts that surround contemporary translation, a picture emerges not only of an industry that has struck a new balance among stakeholders in a moment of technological transition but also of broader shifts in the provision of services and the uses of language.

The logic of the ProZ.com website is familiar: it is functionally similar to sites like oDesk, eLance, and the craftwork clearinghouse Etsy. Those seeking translation services search the database for a translator with the desired profile. Even simpler, they can post job announcements and let the translators come to them. On the other side of the equation, freelancers who create accounts read the listings and make bids. But the website hides most jobs behind a pay wall for 12 hours or longer. In order to get at the work before it dries up, freelancers must surrender a membership fee. From ProZ.com’s perspective, a professional freelance translator is someone who sends US$129 every 12 months.

The language ProZ.com speaks is not translation theory; it’s investment, success, and satisfaction guaranteed. A paid membership is a ‘tried and true means of boosting your business’ (ProZ.com, 2010). Testimonials on the site reinforce this message. Displayed in several languages, testifying to the site’s global reach, they address freelancers seeking to expand beyond regional markets. But perhaps the clearest confirmation of ProZ.com’s impact on the freelance market and the image it projects lies in one member’s claim that ‘almost 100% of [her] business is there thanks to ProZ.com’ (ProZ.com, 2010). As another raved, ‘I became a member of ProZ.com last month and I’ve already made 2.5x more the price of the membership’ (ProZ.com, 2011). The happiness of ProZ.com’s customer is correlated with perceived return on investment. But investments can fail or take time to mature. So ProZ.com hedges: ‘Nothing good ever came easy’ (ProZ.com, 2010).

Once the user pays up, the sales pitch yields to the delivery of the goods. ProZ.com’s challenge is to provide as many of its paying freelancers as possible with enough work to justify the expense of purchasing next year’s membership. The company’s success is measured in membership renewal. To achieve this, the site must attract an adequate amount of paying work. But this means keeping a steady supply of freelancers. In effect, ProZ.com’s business model is a balancing act: to succeed, a sufficient stock of supply and demand must constantly flow into the system.

But the supply and demand that meet in the ProZ.com marketplace must be of the variety that delivers decent translations and pays its bills on time. While it is clear that the company has an interest in an overwhelming supply of freelancers who generate an overwhelming supply of membership fees, ProZ.com has also developed quality control mechanisms to weed out poor translators and outsourcers who stiff freelancers. After a job is complete, outsourcers can indicate the likelihood that they will hire a freelancer again; similarly, freelancers can rate their willingness to work with an agency again. Comments and responses become part of a freelancer or outsourcer’s permanent record. Before accepting a job, a paying member can review comments left by freelancers who have worked for an outsourcer before, effectively running background checks on potential clients. A translation seeker is likewise able to review freelancers’ ratings before offering a job.

To boost higher-quality translators further, ProZ.com also invites freelancers to earn merit points by responding to inquiries on a message board. These so-called KudoZ are tracked and displayed on translators’ profiles – crowdsourced measures of freelancers’ expertise. More recently, paying freelancers can also request membership in the company’s Certified PRO Network. Though there is no charge to freelancers beyond the membership fee (ProZ.com, 2008), ProZ.com devotes staff time to examine credentials issued by professional associations such as the ATA and to review sample translations. Translators who qualify can display their certifications on their profile pages. In parallel with the PRO Network, ProZ.com has also introduced premium packages for outsourcers. Corporate memberships and the ProZ.com Connect platform offer translation seekers granular levels of access to the translator database. Agencies who sign up for these programs are expected to maintain high community ratings. Moreover, by paying US$250 and US$320 annually, corporate members and ProZ.com Connect members are awarded digital badges that distinguish them as trustworthy vendors. ProZ.com’s investments in credentialing are indicative of the value of maintaining quality among translators and agencies in the marketplace. KudoZ, the PRO Network, and premium corporate memberships are algorithmic mechanisms that manage the flood of buyers and sellers on the translation market and mitigate the complaints and concerns that bubble up periodically among experienced translators about plummeting rates and quality (Dam and Zethsen, 2010: 206).⁷

The new translation appears in stark terms: vendors are granted easy access to a global workforce and entrepreneurialism is the tune that ProZ.com’s 300,000 freelancers whistle while they work. The key to making the investment in a membership pay off is not to translate better but to maximize revenue streams: the ProZ.com platform is designed to allow freelancers to subcontract work easily, potentially turning any translator into an agency. ProZ.com is by no means an altruistic website, but its business model is predicated on clients seeing their own profits increase. In this sense, even in its freelance manifestation, where the focus is on ‘scientific, technical, commercial, legal and administrative or institutional’ texts (Cronin, 2003: 1), and even in the age of computer-aided translation tools and translation memory software packages, the globalized translation industry remains very much the realm of the creative and the imaginative.

The dream of automatic translation

In celebrating the creativity that the globalized translation industry demands of freelance labor, we may have mistaken the forest for the trees. The recent management science literature deploys the notion of entrepreneurialism to describe behavior whereby workers incorporate functions reserved to management under Fordism (cf. Neilson and Rossiter, 2008). Echoing Foucault’s (1991) notion of governmentality, which imputes a moral responsibility to pursue ‘the correct manner of managing individuals, goods and wealth’ (Foucault, 1991: 92), the responsibilities of strategy development, accounting, invoicing, benefits administration, space management, procurement, and accounts management are assumed by enterprising freelancers, or workers who have been labeled ‘precarious’.⁸ Ultimately, the issue of precarity in the translation industry is only half interesting. Translators’ precarity is nothing new. Venuti (1995) noted that translators’ wages have historically flirted with the poverty level, creating armies of part-time workers who undertake of a range of other work – he identifies editing, writing, and teaching (Venuti, 1995: 11) – to remain solvent. What matters is how changes in the media environment change the specificity of freelancers’ casualization.

The increased prevalence of freelance labor found by Fraser and Gold (2001) in the late 1990s delivers clear advantages to agencies and clients. By shrugging off contracts that impose salary and other obligations, employers reduce costs (cf. Houseman, 2001: 155–157). As one freelance journalist wrote (in an article fittingly assigned by a freelance editor), freelancers are ‘paid for [their] output rather than [their] time’ (Clapperton, 2010: 69). But translation seekers realize another advantage. Beyond shifting the financial obligations associated with retaining translators, the rise of entrepreneurialism in the industry has allowed buyers to imagine an anonymous, automated translation factory that operates at all hours in all language combinations on short deadlines. Beyond Google Translate and experiments with semi-skilled microlabor, consider ProZ.com’s recent launch of a ‘turnkey translation’ service that promises ‘professional-quality translation of a short text without having to search out qualified translators or handle project management’ (ProZ.com, 2012). Under this scheme, ProZ.com takes the source text, collects pre-payment at fixed rates (including a transaction fee), matches the text algorithmically to a translator, delivers the translation, and issues payment to the translator. Poor reviews of translators’ work exclude them from future turnkey projects.

ProZ.com thus emerges as a temporary stand-in for the ultimate translation dream: friction-free machine translation. In 2009, the Obama White House stressed the economic importance of developing ‘[a]utomatic, highly accurate and real-time translation between the major languages of the world’ (National Economic Council, 2009). BusinessWeek quickly observed that machine translation has long been ‘the next big thing’ and reported that humans remain necessary for the production of usable translations (Joseph, 2009). Notwithstanding recent advances, reliable machine translation hovers on the always-receding horizon. ProZ.com offers real clients the possibility of obtaining translations quickly, easily, and cheaply. Submission of a simple web form (which language pairs? how many words? by what deadline?) returns bids from eager freelancers, whose credentials can be examined and whose entrepreneurial spirit trains them to keep rates competitive. For small jobs, the turnkey option delivers human-quality results. For more complicated projects or for end clients who wish not to be bothered, an agency can be retained to operate the ProZ.com interface, vet the freelancers, and perform quality control. In the globalized translation industry, existing relationships are no longer the limiting agent in the translation reaction – specify the appropriate parameters and anything can be translated.

Built on a set of relational databases (providers, seekers, agencies, listings, membership levels) that return results through a packet-switched distributed network, ProZ.com represents the necessary middle layer that transforms the industry from a relatively linear model to one that mimics the infrastructure by which the ProZ.com website itself is circulated. This mimicry operates on two levels. First, the internet is among the most familiar manifestations of distributed, many-to-many networks. Similarly, the globalized translation industry connects clients with translators who are geographically distributed and work in various language combinations. More profoundly, the globalized translation industry mimics the internet’s overarching logic. The packet-switching operation that moves data from one network node to another depends upon highly stylized metadata. The internet is wholly uninterested in the content a user summons; its routers and browsers care only that headers properly describe the size and type of the data sought. It is, as Galloway writes (2004), a universe governed by protocols where breaches of conduct are punished by denials of service. So long as the languages and deadline are properly specified and the fees paid, ProZ.com will enable translation seekers to find translation providers. ProZ.com is no more interested in a translation project’s contents than a barge captain is in the contents of the shipping containers piled on his deck (cf. Levinson, 2006; Morley, 2011).

The translation industry may be said to pursue what Naoki Sakai (2009) calls ‘the idea of the unity of language’, which makes it possible ‘to systematically organize knowledge about languages in a modern, scientific manner’ (Sakai, 2009: 73). Such an idea is essential for any standardized, automated, algorithmic approach to translation. The smooth functioning of the translation industry under globalization demands conceptual containers (‘unified languages’) just as transoceanic transport requires uniform containers. Indeed, freelance translation necessitates the adoption of what Sakai (1997) elsewhere describes as the ‘homolingual address’, which ‘assumes the normalcy of reciprocal and transparent communication’ (Sakai, 1997: 8). Under the cover of homolingualism, the algorithm can deal in fixed commodities (‘English’, ‘French’, ‘pharmaceuticals’) while humans are relegated to teasing out the finer points of language and its social wrappings: a translation’s exchange value eclipses its use value (Basalamah, 2008: 359–360). These social wrappings are the stuff of Sakai’s (1997) ‘heterolingual address’ (Sakai, 1997: 7ff). Even if ‘translation involves bodies’ that are able to translate ‘not texts, but contexts’ (Ivekovic, 2001–2002: 123), the most profound effect of the freelance translation machine is to develop an interface connecting (and simultaneously separating) the homolingual and the heterolingual, the machine and the human.

In this sense, translation has been partially undifferentiated from other types of linguistic services (technical writing, copywriting, editing), creative craftworking, and even, by extension, social functions such as party coordination, birth announcements, and friendship. The freelance translation machine is less a matter of inventing a new way of delivering translation than devising an algorithm that enlists entrepreneurialism to motivate freelance workers to act as an extra-dimensional do loop and produce creative products within a standardized framework. Whether the specific example is ProZ.com’s automation of translation, Etsy’s rationalization of arts and crafts, or Facebook’s monetization of sociality, the encounter of flesh and machine marks a moment where algorithms prove unable to complete a task and the human mind is inserted to complete the circuit of efficient and frictionless commerce.⁹

The freelance translation machine described here represents a kinder, gentler instance of algorithmic culture. The interests of capital are well served by online freelance labor marketplaces that cut across geographic boundaries. But the inclusion of human cognition in an otherwise computational system allows for multiple winners. Thinking of video games, Galloway (2006) writes:

[T]he gamer is not simply playing… The gamer is instead learning, internalizing, and becoming intimate with a massive, multipart, global algorithm. To play the game means to play the code of the game. To win means to know the system. And thus to interpret a game means to interpret its algorithm. (Galloway, 2006: 90–91).

The freelancers who become ProZ.com users can instrumentalize the interface even as it instrumentalizes them: by internalizing the logics of entrepreneurialism and learning to operate the ProZ.com interface, they ‘introduce economy’ (Foucault, 1991: 92) into their operations, increasing productivity to the benefit of the industry and extracting some degree of compensation. Those who produce quality translations, receive high ratings, develop relationships with outsourcers and come to attract new clients will easily justify next year’s US$129 membership fee. Their motivations are many: some seek to earn supplemental income, others pursue full employment, and still others set up agencies of their own that are little more than email inboxes. Some move in and out of these and other roles as their needs change over time. Freelancing today is no more or less precarious than in previous media regimes: just as before, some win while others lose. The difference now is that success in freelancing is as much a matter of mastering algorithms as of achieving linguistic excellence or grooming social relationships.

Mastery of the algorithm does not necessarily lead to its overthrow. It can instead mark an uneasy peace along the borderlands separating humans and machines. Such a critique of algorithmic culture is thus somewhat more ambivalent than that of Galloway or Striphas. The scepter of Galloway’s protological control is blunted by the possibility of orthogonal human goals. Meanwhile, Striphas’s concerns with the use of computation to measure and shape tastes via black boxes are complicated by the possibility that algorithms are agnostic toward cultural content and interested only in metadata. Enlisting the human mind as a subroutine yields a do loop that is simultaneously extra-dimensional, in that it explodes out of binary computation, and leaky, in that human laborers are not necessarily bound by the same motivations as their algorithmic counterparts.

Footnotes

Acknowledgements

This article benefitted from the thoughtful critiques of two anonymous reviewers, conversations with Christine Mitchell and Marc Charron, and generous audiences at Princeton University and the University of Ottawa at the invitations of David Bellos and Marc Charron, respectively. Philip J. Smith offered guidance on commerce in antiquity, Jan Jörgensen explained measures of market size, and Nathalie Cooke helped make connections. All errors and infelicities are my own.

Funding

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Notes

Author biography

Scott Kushner studies the paradoxical interface between analog humans and digital machines. He completed a PhD in French at Duke in 2009, taught in American Studies at Rutgers, and currently holds an administrative position at McGill.

References

Acland

(2003) Screen Traffic: Movies, Multiplexes, and Global Culture. Durham, NC: Duke University Press.

American Translators Association (2003) Translation and Interpreting: Compensation Survey. Alexandria, VA: ATA.

Aneesh

(2006) Virtual Migration: The Programming of Globalization. Durham, NC: Duke University Press.

aquarius.net (2010) Aquarius.net. Available at: www.aquarius.net (accessed 2 June 2012).

Basalamah

(2008) Le Droit de Traduire: Une Politique Culturelle pour la Mondialisation. Ottawa, ON, Canada: Presses de l’Université d’Ottawa.

Beninatto

(2006) Translator portal ProZ.com aims for enterprise. Global Watchtower, 30 October. Available at: http://upgrade.globalwatchtower.com/2006/10/translator-portal-ProZ.com-enterprise (accessed 2 June 2012).

Boucau

(2005) The European translation industry: Facing the future. Available at: www.guilde.net/fr/news/euatc.pps (accessed 20 May 2012).

Chabert

J-L

Barbin

Borowczyk

. (1999) A History of Algorithms: From the Pebble to the Microchip. Berlin: Springer.

Charron

(2005) Plus vite, encore plus vite: La traduction à l’heure de la mondialisation. Translation Studies in the New Millennium 3: 15–27.

10.

Clapperton

(2010) How technology will change the way we manage. Engineering and Technology 5(1): 66–69.

11.

Cronin

(2003) Translation and Globalization. London: Routledge.

12.

Dam

Zethsen

(2010) Translator status: Helpers and opponents in the ongoing battle of an emerging profession. Target 22(2): 194–211.

13.

DiTomaso

(2001) The loose coupling of jobs: The subcontracting of everyone. In: Berg

Kalleberg

(eds) Sourcebook of Labor Markets: Evolving Structures and Processes. New York: Kluwer, pp.247–270.

14.

Fiederer

O’Brien

(2009) Quality and machine translation: A realistic objective? Journal of Specialised Translation 11: 52–74.

15.

Foucault

(1991) Governmentality. In: Burchell

Gordon

Miller

(eds) The Foucault Effect: Studies in Governmentality. Chicago, IL: University of Chicago Press, pp.87–104.

16.

Fraser

Gold

(2001) ‘Portfolio workers’: Autonomy and control amongst freelance translators. Work Employment and Society 15: 679–697.

17.

Galloway

(2004) Protocol: How Control Exists after Decentralization. Cambridge, MA: MIT Press.

18.

Galloway

(2006) Gaming: Essays on Algorithmic Culture. Minneapolis, MN: University of Minnesota Press.

19.

Galloway

(2008) The unworkable interface. New Literary History 39(4): 931–955.

20.

Gehl

(2011) The archive and the processor: The internal logic of Web 2.0. New Media & Society 13(8): 1228–1244.

21.

Gill

Pratt

(2008) In the social factory? Immaterial labour, precariousness and cultural work. Theory, Culture & Society 25(7–8): 1–30.

22.

Global Voices Online (2012) Lingua volunteer FAQ. Available at: http://globalvoicesonline.org/lingua/lingua-volunteer-faq/ (accessed 3 October 2012).

23.

Glotz

(1965 [1926]) Ancient Greece at Work: An Economic History of Greece from the Homeric Period to the Roman Conquest (trans. Dobie

). New York: Barnes & Noble.

24.

Gold

Fraser

(2002) Managing self-management: Successful transitions to portfolio careers. Work Employment and Society 16: 579–797.

25.

Google (2012) Find out how our translations are created. Available at: http://translate.google.com/about/intl/en_ALL (accessed 24 May 2012).

26.

Gouadec

(2007) Translation as a Profession. Amsterdam: John Benjamins.

27.

Hardt

Negri

(2000) Empire. Cambridge, MA: Harvard University Press.

28.

Hardt

Negri

(2004) Multitude: War and Democracy in the Age of Empire. New York: Penguin.

29.

Hardt

Negri

(2009) Commonwealth. Cambridge, MA: Harvard University Press.

30.

Hayles

(1999) How We Became Posthuman: Virtual Bodies in Cybernetics, Literature, and Informatics. Chicago, IL: University of Chicago Press.

31.

Hermans

Lambert

(1998) From translation markets to language management: The implications of translation services. Target 10(1): 113–132.

32.

Hosaka

(2008) Facebook asks users to translate for free. MSNBC.com, 18 April. Available at: www.msnbc.msn.com/id/24205912/ns/technology_and_science-internet (accessed 21 May 2012).

33.

Houseman

(2001) Why employers use flexible staffing arrangements: Evidence from an establishment survey. Industrial & Labor Relations Review 55(1): 149–170.

34.

Ivekovic

(2001–2002) De la traduction permanente (Nous sommes en traduction)/On permanent translation (We are in translation). Transeuropéennes 22: 121–145.

35.

Joseph

(2009) White House challenges translation industry to innovate. Bloomberg Businessweek, 1 October. Available at: www.businessweek.com/innovate/content/oct2009/id2009101_196515.htm (accessed 21 August 2011).

36.

Kay

(2003) The proper place of men and machines in machine translation. In: Nirenburg

Somers

Wilks

(eds) Readings in Machine Translation. Cambridge, MA: MIT Press, pp.221–232.

37.

Levinson

(2006) The Box: How the Shipping Container Made the World Smaller and the World Economy Bigger. Princeton, NJ: Princeton University Press.

38.

Meedan (2012) Meedan. Available at: http://news.meedan.net/ (accessed 3 October 2012).

39.

Mitchell

(2010) Situation normal, all FAHQT’d up: Language, materiality & machine translation. PhD Dissertation, McGill University, Montreal, QC, Canada.

40.

Morley

(2011) Communications and transport: The mobility of information, people and commodities. Media Culture & Society 33: 743–759.

41.

Mossop

(2000) The workplace procedures of professional translators. In: Chesterman

Gallardo San Salvador

Gambier

(eds) Translation in Context: Selected Contributions from the EST Congress, Granada 1998. Amsterdam: John Benjamins, pp.39–48.

42.

Mossop

(2006) Has computerization changed translation? Meta 51: 787–805.

43.

National Economic Council (2009) A strategy for American innovation: Driving towards sustainable growth and quality jobs. Available at: www.whitehouse.gov/administration/eop/nec/StrategyforAmericanInnovation (accessed 21 August 2011).

44.

Neilson

Rossiter

(2008) Precarity as a political concept, or, Fordism as exception. Theory Culture & Society 25(7–8): 51–72.

45.

Perrino

(2009) User-generated translation: The future of translation in a Web 2.0 environment. Journal of Specialised Translation 12: 55–78.

46.

ProZ.com (2008) Certified translators. Available at: www.ProZ.com/pro-tag/info/about/freelancers (accessed 20 May 2012).

47.

ProZ.com (2010) Invest in success. Available at: www.ProZ.com/membership/campaign (accessed 15 November 2010).

48.

ProZ.com (2011) Testimonials and success stories. Available at: www.ProZ.com/testimonials (accessed 15 September 2011).

49.

ProZ.com (2012) ProZ.com turnkey translation. Available at: www.ProZ.com/order-translation (accessed 20 May 2012).

50.

Pym

(1995) Translation as a transaction cost. Meta 40: 594–605.

51.

Pym

(2004) The Moving Text: Localization, Translation, and Distribution. Amsterdam: John Benjamins.

52.

Pym

(2006) Globalization and the politics of translation studies. Meta 51: 744–757.

53.

Racette

(2007) Exploring freelance business practices. ATA Chronicle, October. Available at: www.atanet.org/chronicle/feature_article_october2007_freelance.php (accessed 2 June 2012).

54.

Robinson

(1997) Translation as repayment of debt. Delos 7(1–2): 10–22.

55.

Romaine

Richardson

(2009) State of the translation industry 2009: Smarter, more casual. Available at: www.slideshare.net/dmc500hats/mygengocom-state-of-global-translation-industry-2009 (accessed 2 June 2012).

56.

Ross

(2002) No-Collar: The Humane Workplace and Its Hidden Costs. New York: Basic Books.

57.

Ross

(2004) Low Pay, High Profile: The Global Push for Fair Labor. New York: Free Press.

58.

Ross

(2006) Fast Boat to China: Corporate Flight and the Consequences of Free Trade – Lessons from Shanghai. New York: Pantheon.

59.

Ross

(2008) The new geography of work: Power to the precarious? Theory, Culture & Society 25(7–8): 31–49.

60.

Sakai

(1997) Translation and Subjectivity: On ‘Japan’ and Cultural Nationalism. Minneapolis, MN: University of Minnesota Press.

61.

Sakai

(2009) How do we count a language? Translation and discontinuity. Translation Studies 2(1): 71–88.

62.

Scholz

(2009) On MTurk, some examples of exploitation. Esse, nosse, posse: common wealth for common people. Available at: http://www.emst.gr/commonwealth/?p=293&lang=el (accessed 3 December 2012).

63.

Search Engine People (2011) 10 inexplicable Google Translate fails. Search Engine People. Available at: www.searchenginepeople.com/blog/10-google-translate-fails.html (accessed 20 May 2012).

64.

Striphas

(2010) How to have culture in an algorithmic age. The Late Age of Print. Available at: www.thelateageofprint.org/2010/06/14/how-to-have-culture-in-an-algorithmic-age (accessed 17 May 2012).

65.

Striphas

(2011) Algorithmic culture, redux. The Late Age of Print. Available at: www.thelateageofprint.org/2011/03/15/algorithmic-culture-redux/ (accessed 17 May 2012).

66.

Toury

(1995) Descriptive Translation Studies and Beyond. Amsterdam: John Benjamins.

67.

Venuti

(1995) The Translator’s Invisibility: A History of Translation. London: Routledge.

68.

Weaver

(2003) Translation. In: Nirenburg

Somers

Wilks

(eds) Readings in Machine Translation. Cambridge, MA: MIT Press, pp.13–17.

69.

Wooten

(2008) How large is the translation industry? Well, how do you define it? T & I Business. Available at: http://tandibusiness.blogspot.ca/2005/12/size-of-t9n-l10n-industry-based-on-how.html (accessed 2 June 2012).

70.

Zaidan

Callison-Burch

(2011) Crowdsourcing translation: Professional quality from non-professionals. In: Proceedings of the 49th annual meeting of the Association for Computational Linguistics, Portland, OR, pp. 1220–1229.