Translation as discrimination: Sociolinguistics and inequality in multilingual institutional contexts

Abstract

Sociolinguistic approaches to social justice tend to treat the use of interpreters or translators as a remedy to linguistic inequality in multilingual institutional settings. This article challenges this assumption by showing how translation can instead contribute to inequality and discrimination. Drawing on studies of face-to-face interpreting in judicial contexts and of written translation in linguistic landscapes, it explores inequalities found in habitual practices of professional interpreters and in the use of machine translation. It shows how language ideologies about multilingualism motivate translation practices that systematically restrict the participation of speakers of subordinated languages, or that stereotype them as deviant when addressed solely by prohibitions and warnings, a practice I call ‘punitive multilingualism’. The article thus argues that sociolinguistic studies of multilingualism should pay closer attention to translation practices within a wider context of language contact and in relation to phenomena such as translanguaging, mock languages, or language shift. (Translation, interpreting, justice, linguistic landscape, discrimination)*

INTRODUCTION

In recent years, sociolinguists and linguistic anthropologists have increasingly turned towards exploring how linguistic diversity relates to justice and what role language plays in processes of discrimination and exclusion (see e.g. Piller 2016; Rickford & King 2016; Baugh 2018; Avineri, Graham, Johnson, Conley Riner, & Rosa 2019). This is arguably a central question for our discipline, and at its core is a central paradox for sociolinguistics. On the one hand, linguistics as a whole is committed to what DeGraff (2020:e300) has called ‘the core universalist-uniformitarian-egalitarian credo’, namely that, in the words of Sapir (1933:155), all languages are ‘an essentially perfect means of expression and communication’, and consequently, are equally worthy of being spoken. On the other hand, as noted by Duranti (2011) language is a ‘non-neutral medium’ that serves as a system of social differentiation. As a result, languages are not simply interchangeable in the social lives of their speakers, and as observed by Blommaert (2003:615), linguistic difference ‘quickly and quite systematically’ gives rise to inequality between speakers. This kind of inequality is apparent especially in contexts where agents of a governmental institution use a particular, often officially sanctioned language, and interact with individuals who speak differently. For example, we find inequality in the judicial sphere when individuals do not understand their rights as told by the police (Eades & Pavlenko 2016), or when individuals do not understand accusations made against them in court (Haviland 2003), or when they are not believed because their dialect is ‘misheard and maligned’ (Rickford & King 2016). Inequality is also prevalent in many other institutional contexts, such as for example health care, when patients’ narratives are not heard and understood by doctors (Briggs 2017), or when potentially life-saving information, such as on how to avoid infection with COVID-19, is communicated only to individuals who are literate in the official language (Piller, Zhang, & Li 2020).

Where sociolinguists address such issues of linguistic inequality, the use of interpreters or translators is often viewed as a remedy, especially in the field of language and law. For example, regarding the communication of rights to non-native speakers in police custody, an international group of linguists, psychologists, lawyers, and interpreters has recently developed a set of guidelines, which have been endorsed by several professional organizations in linguistics and applied linguistics (Communication of Rights Group 2016). These guidelines advocate both for the use of written translation and professional interpreting, recommending among other things that ‘all vital documents must be made available in a language the suspect can understand’ and that ‘all non-native English-speaking suspects should be provided with the opportunity to request the services of a professional interpreter for the police interview’. Similarly, addressing the discrimination of speakers of African American Vernacular English in US courts, Rickford & King (2016:981) propose that linguists ‘[a]dvocate for the use of interpreters as an option for deep AAVE and other vernacular speakers’. Translation is also endorsed in linguistic landscape studies, albeit often only implicitly, by treating the parallel presence of multiple languages as more equitable than the exclusive use of a single dominant language. For example, Landry & Bourhis (1997:27) argue that the presence of a language in the linguistic landscape ‘can contribute most directly to the positive social identity of ethnolinguistic groups’, whereas the exclusion of a language from public signage may ‘convey… that one's own language is not valued and has little status within society… reinforcing a diglossic situation to the advantage of the dominant language’ (Landry & Bourhis 1997:28). Their arguments have been taken up by other scholars concerned with language rights and inequality, who have treated the absence or exclusion of relevant languages from the linguistic landscape as a discriminatory mechanism of a language policy (e.g. Shohamy 2006:123; Marten, van Mensel, & Gorter 2012; Rubdy & Ben Said 2015). By contrast, when governments do provide parallel texts in multiple languages, this generally involves processes of translation, but these are typically not discussed as such in linguistic landscape studies.

Translation is not a panacea, however. While it can be seen as a tool for reducing inequality, specific translation practices in institutional contexts may in fact contribute to injustices as well, as has been demonstrated in some studies, particularly in work on interpreting in refugee contexts (e.g. Jacquemet 2005; Blommaert 2009; Inghilleri 2012; Maryns 2013; Haviland 2019). In this article, I examine discriminatory practices in spoken interpreting and in written translation, and caution that we cannot assume that translation—or more, generally, the parallel availability of multiple languages—automatically leads to inclusion and to something we might want to call linguistic justice. On the contrary, it can contribute to inequality and discrimination, for example, when interpreting is provided in ways that prioritize the needs of the institution over that of the speakers of other languages who are ostensibly served by it. Moreover, institutions may limit their use of other languages to contexts where they issue warnings or threaten punishments for deviant behaviour, a practice for which I propose the term punitive multilingualism. As I aim to show in this article, institutions may create an appearance of inclusivity and open-mindedness when they provide interpreting or produce multilingual written signs. However, in a way that is reminiscent of Hill's (1998) observations about the dual indexicality of Mock Spanish, institutions may at the same time discriminate against other-language speakers, when these translation practices create or reinforce stereotypes or disregard the actual communicative needs of lay participant. As a consequence, the questions—what gets translated, why, how, and by whom—are central to the evaluation of linguistic practices in multilingual institutional contexts. Examining translation and interpreting more closely and systematically is therefore essential for sociolinguists and linguistic anthropologists who seek to use their scholarship to further social justice in multilingual societies.

SOCIOLINGUISTICS AND TRANSLATION

Translation and interpreting have largely been absent from theoretical debates in sociolinguistics, or even linguistics more generally. For example, recent volumes that survey the field do not include chapters on translation or interpreting, and some do not even list these terms in their index (see e.g. Mesthrie 2011; Wodak, Johnstone, & Kerswill 2011; Bayley, Cameron, & Lucas 2013; Coupland 2016). Sociolinguistic journals do occasionally publish articles on translation, but there have not been many systematic attempts to develop a sociolinguistic theory of translation, or to engage with the field of translation studies. One exception to this trend involves explorations of the translation of social meaning, such as the stylistic representation of speakers of African American Vernacular English in translations into German or Japanese (Berthele 2000; Inoue 2003; Queen 2004; Hiramoto 2009). The difficulty of translating indexical meaning has also been discussed by Blommaert (2006:173), Gal (2015), and especially Silverstein (2003:87) who proposed to use the term transduction to describe how ‘source-text indexical values have to be reconstructed in indexical systems of another culture’, while reserving the term translation more narrowly for correspondences in denotational meaning. This distinction between referential and indexical meaning is also central to Haviland's (2003) analysis of interpreting practices in US courts, which he shows to be rooted in language ideologies that also shape judicial decision-making about multilingualism and communication. Specifically, he identifies a prevalent ideology of ‘referential transparency’ in which language is reduced to the communication of propositional information, and words in another language are merely an ‘exotic costume for a shared meaning’ (Haviland 2003:772). Haviland also shows how the institutional assignment of translation is rooted in standard language ideology that categorizes (and often miscategorizes) individuals as speakers of particular, named languages, irrespective of their actual communicative repertoires.

Overall, linguistic anthropologists have tended to engage with translation much more than sociolinguists, for example, by exploring the role of translation in furthering cultural and linguistic change, such as in the context of Christian missionization and the development of new Christian registers of indigenous languages in Papua New Guinea or the Americas (Schieffelin 2007; Hanks 2010, 2015; Mannheim 2015). Hanks (2010:38) argues that this involves a process of commensuration, that brings previously ‘incommensurable cultural worlds’ into sufficient alignment so as to make translation possible. This process is marked by a power asymmetry in which ‘it is the subordinate language that is altered’ (Hanks 2010:38), showing how translation is often deeply implicated in inequality between linguistic communities in contact, and as part of processes that ‘create or manage marginalized, subordinated populations’ (Gal 2015:227). Power asymmetries in translation have also been addressed by scholars who have examined interpreter-mediated interaction using approaches influenced by conversation analysis and Goffman's participation framework. Such studies have investigated the communicative practices of interpreters and their impact on the interactions in which they participate, as well as the social identity of interpreters as intermediaries in language contact (Davidson 2000; Berk-Seligson 2009; Angermeyer 2009; Vigouroux 2010). This is arguably the area where the encounter between sociolinguistics and translation studies has been the closest, though findings from such research have not always had much uptake in the wider field of sociolinguistics.

In this article, I build on this work to investigate the participation framework of spoken and written translation in institutional contexts, asking whose voice is translated and how, but also who is addressed, and who is an unaddressed recipient. Based on the participation framework and the direction of translation, two basic situations can be distinguished. An individual whose language variety differs from that used in an institution can be a speaker or a hearer, and their language (which can be characterized as a subordinated variety) can either be the source language, for translation into the institutional language, or it can be the target language, for translation from the institutional language. In the first case, the speaker of the subordinated language is the principal, and it is the task of the translator or interpreter to give them a voice in the dominant institutional language context. In the second case, a speaker of the subordinated language is the intended recipient of translated speech or written text. The principal here is the institution or its representatives, and the translator's task is to communicate their position to the lay participant, ranging from ‘giving information’ to ‘giving orders’. In both cases, translators or interpreters act primarily as animators and authors, though not always exclusively.

In institutional contexts, written translation is typically very asymmetrical in its directionality, with the subordinated language as the target rather than the source. Interpreting, by contrast, is more often bidirectional, but the two directions are not necessarily treated the same way; for example, they may involve different modes of interpreting. This article focuses on instances of translation where the subordinated language is the target, beginning with observations about court interpreting, before turning to written translation in the linguistic landscape.

INTERPRETING IN INSTITUTIONAL INTERACTION

Linguists’ calls to make institutional encounters more just for individuals with limited or no proficiency in the institutional language have often focused on the provision of accurate and impartial interpreting. In particular, some scholars have argued in favor of using professional interpreters with specialized training and accreditation in contexts ranging from the courtroom (Berk-Seligson 1990; Hale 2004) or asylum proceedings (Maryns 2013), to interactions with police (Pavlenko 2008; Berk-Seligson 2009; Nakane 2014; Eades & Pavlenko 2016), or with healthcare providers (Angelelli 2004; Cox & Maryns 2021). These efforts serve to address and avoid demonstrated problems with ad hoc, non-professional interpreting, particularly with regard to perceived translation accuracy and completeness. However, such professionalization of interpreting tends to be available only for certain standardized languages, leading to inequality between different subordinated varieties and their speakers. Furthermore, even with professional interpreters, interpreting in institutional contexts is frequently characterized by asymmetry between the participants, which affects their ability to understand and make themselves understood, their ability to take up certain participant roles, and the way in which their own language use is monitored and accommodated to by the interpreter.

Interpreting modes

As shown in Angermeyer (2015), one source of asymmetry is in the distribution of consecutive and simultaneous interpreting modes, and this is particularly common in court interpreting. Interpreting from the subordinated language into the language of the institution is typically done in short consecutive mode, where speakers have to pause for the interpreter at regular intervals. This happens, for example, when a witness testifies in another language, responding to questions posed by a judge or attorney. This mode is also common in other institutional contexts such as police interviews or doctor-patient interviews that involve comparable patterns of questions and responses. In such situations, consecutive interpreting emerges interactionally because participants generally need to wait for the interpreter to translate what their interlocutor has said.¹

This differs from a situation in which interaction between several speakers of the dominant language is translated for the benefit of a third person who speaks a different language. This is common practice in many jurisdictions when the interpreter translates the proceedings for a defendant who does not understand the language of the court. Sometimes called proceedings interpretation, this practice is intended to protect a defendant's legal right of due process, that is, their ability to understand the charges and participate in their defense against them. As the primary interlocutors in this situation do not require interpretation to understand one another, they typically do not orient towards the task of interpreting and do not pause to facilitate it. As a consequence, interpreters are forced to interpret in simultaneous mode, speaking at the same time as they are listening to new source input. This is a frequent practice in court interpreting, where it is often referred to as whispering mode (chuchotage), when the interpreter speaks in a quiet voice that is audible only to persons nearby.

This asymmetrical distribution of interpreting modes is extremely common; it has been described for courts in many jurisdictions worldwide (see e.g. Berk-Seligson 1990; Kadric 1999; Jacobsen 2012; Ng 2018). In Angermeyer (2015), I have argued that it inherently disadvantages speakers of subordinated languages in two ways, both when they speak and when they listen to others. As speakers whose speech is interpreted in consecutive mode, they are more likely to be interrupted than other participants are, because when they pause for the interpreter, these pauses may be mistaken for transition relevant places (see D'hondt 2004). Even without such interruptions, the pauses cause narrative testimony to be more fragmented and less compelling or credible (Angermeyer 2021). Conversely, as hearers who depend on the interpreter for information about the proceedings, speakers of subordinated languages are disadvantaged by the use of simultaneous interpreting because it is more likely to lead to omissions than short consecutive interpreting does. This is the case because simultaneous interpreting is cognitively more demanding for interpreters (Pöchhacker 2004:100; Seeber 2011). Interpreters frequently struggle to keep up with the simultaneous source talk, particularly if it involves overlapping turns from multiple speakers, and this may cause portions of source talk to remain untranslated. The simultaneous mode also makes it much more difficult for the listener to check their own understanding. These disadvantages do not arise for speakers of the institutional language. When they speak, they do not need to pause for the interpreter and are thus less likely to experience interruptions and narrative fragmentation. When they listen to the interpreter, they do so in short consecutive mode and are thus more likely to hear a close rendition, and are better able to check their own comprehension.

Participant roles

A further asymmetry in court interpreting is in the distribution of participant roles. As noted, the practice of chuchotage in proceedings interpreting makes it difficult for both the interpreter and the recipient of the interpretation to check their understanding. This is the case because, at that moment, they are not ratified participants and are not able to take the floor as speakers. In court interpreting, this limitation of the participant roles is reflected in conventions about the use of person deixis by interpreters. In many jurisdictions, interpreters are explicitly required by the institution to maintain the person deixis of the source utterance in their target rendition. That is, first person and second person forms in the target rendition refer to the speaker and addressee of the source, respectively. During proceedings interpretation, this means that interpreters use second person address forms that do not address the lay participant who is listening to them, but instead are meant to refer to another participant who was addressed in the source. This may lead to misunderstandings, because recipients are likely to feel addressed, as shown in example (1), from an arbitration hearing about a housing dispute with a Polish-speaking claimant (see Angermeyer 2015:86).

In line 1, the arbitrator uses you to address the English-speaking defendant, and she refers to the claimant in the third person, as this lady. In line 3, the interpreter translates this question into Polish using the same pronominal deixis as in the source: a polite address form for the defendant, and third person for the claimant to whom she is speaking. The claimant's response in line 4 shows that she has taken this to mean that the question is addressed to her, rather than to the defendant. As the sole recipient of the interpreter's Polish speech, the claimant feels addressed by the polite address form ma Pani, and she assumes that the third person z tą panią ‘with this lady’ refers to the defendant rather than to herself. The claimant thus responds, but is immediately interrupted by the interpreter, with the words Ja tylko tłumaczę ‘I'm only translating’. This illustrates a language ideology that conceptualizes translation as a mere word-matching exercise, what Haviland (2003) has termed the ideology of referential transparency. But what the interpreter calls ‘only translating’ can also be viewed as an insistence on an institutional register practice that does not serve the communicative needs of the lay recipient. Crucially, it does not treat the recipient of the interpretation as the addressee, even when they are the only person in the room who can understand the language.² In this example, the Polish speaking claimant is not being talked to in Polish, but rather she is being talked about to her face.³

While this practice may lead to misunderstandings, as shown, it also has the effect of positioning the other-language speaker solely as a passive recipient of the interpreter's rendition, rather than as a ratified participant who may be expected to take the speaking role in a subsequent turn. In this, direct translation can be seen as playing a part in maintaining the institutionally pre-allocated turn distribution, as lay participants are not expected to take turns when they are not explicitly addressed. When lay participants do speak, as the claimant does in example (1), interpreters often refuse to translate their speech, and instead reprimand them for speaking ‘out of turn’. In so doing, interpreters become gatekeepers who monitor and limit the participation of lay participants (Davidson 2000). While interpreters in face-to-face interaction are arguably always involved in coordinating the talk of other participants (Wadensjö 1998; Baraldi & Gavioli 2012), professional interpreters tend to do so in ways that prioritize the needs of the institution that employs them. Speakers of subordinated languages thus depend on the interpreter's goodwill and cooperation to make themselves heard, making resistance to institutional practices and policies much harder to achieve. This shows the potential for interpreting to take on an asymmetrical unidirectional quality, where the primary purpose for the interpreter is to convey institutional instructions to the lay participant.

Such discursive asymmetries have long been described in interpreting contexts that are marked by power asymmetries between different social groups, particularly in the context of colonialism. In Black skin, white masks, Frantz Fanon (1952:17, 2008:3) observed that ‘[i]n the colonial army… the ‘native’ officers are mainly interpreters. They serve to convey to their fellow soldiers the master's orders, and they themselves enjoy a certain status’. A similar dynamic can be observed in the Inuktitut-language feature film One Day in the Life of Noah Piugattuk, which shows a dramatized account of an interpreter-mediated colonial encounter in the Arctic (Kunuk 2019). Noah Piugattuk was an Inuk hunter who lived from 1900 to 1996, and the film is based on his recollection of an encounter with a Canadian government agent who pressured him and his family to abandon their traditional semi-nomadic lifestyle and move to a permanent settlement controlled by the Canadian state. At the beginning of the encounter, the Inuk interpreter who arrives with the agent explains that he is there so that ‘people can understand what he [the agent] says’, illustrating the communicative asymmetry of the encounter: the colonial agent is there to control the Inuit population and to get them to follow his orders; he is not there to listen to them or learn from them.

These examples illustrate also how face-to-face interpreting is typically done by people who are native speakers of the subordinated language, as speakers of the dominant language do not tend to acquire subordinated languages. This may place interpreters in a precarious position, where their allegiance to the subordinated community conflicts with their service to the institution, and they may end up being distrusted by one of the sides, or both (see e.g. Collins & Slembrouck 2006). Translation and interpreting can take the place of more egalitarian interaction that would involve mutual accommodation, second language acquisition, translanguaging, or the use of a contact language (such as the so-called Eskimo Pidgin, see van der Voort 1995). The power asymmetry of such encounters ultimately sets up a situation where the acquisition of the dominant language becomes necessity for any sustained effort of resistance, any effort to regain agency and a degree of self-control that is not dependent on the cooperation of an interpreter. In other words, it can be seen as a factor driving language shift.

In interpreter-mediated institutional interaction, the reluctance of the interpreter to translate when individuals speak out of turn or in ways that are face-threatening, may prompt participants to switch to the institutional language to the extent that they can. In research in New York small claims courts, I found that litigants who spoke another language often switched to English, whether to answer a yes/no question, to use certain isolated English words, or for more elaborate narrative responses. However, these kinds of codeswitching and codemixing were systematically discouraged by the court, as litigants were told by legal professionals and by interpreters to speak in the other language instead (Angermeyer 2008, 2015). Such directives can be seen again as rooted in a referentialist view of language (Haviland 2003), where the task of a speaker is to ‘put their thoughts into words’, which can then be translated without any loss of meaning into the institutional language. The (usually implicit) assumption is that this putting-thoughts-into-words is better done in one's L1, and by extension, second language speakers cannot be reliably trusted to choose the right words. This frames bilingualism as an exceptional quality and positions interpreters as the only ones able to reliably use both of the so-demarcated languages.

But this view ignores the fact that interpreter-mediated interaction can also be seen as a form of bilingual discourse. In fact, litigants in New York small claims court whose participation is mediated by an interpreter nonetheless listen to what is said in English by other participants, and they often react to it directly, instead of communicating exclusively with the interpreter. In other words, speakers are translanguaging, and are drawing on the linguistic repertoire that they have acquired in their lived experience of language contact. But interpreters often reject this and instead emphasize boundaries between the languages, speaking in a formal, legal register of the subordinated language, that may differ substantially from the vernacular of the lay participants. Translation in institutional contexts can thus enter into a competition with translanguaging and undermine the efforts by lay participants to establish common ground with institutional agents, such as by using terminology from the institutional language. In such situations, lay participants may come to be seen as inadequate speakers of their own vernacular (i.e. as ‘languageless’ following Rosa 2019). Alternatively, and perhaps more commonly, translation hides these sociolinguistic facts from the institution and masks linguistic differences, communication difficulties, or irritations that may arise between interpreters and lay participants.

The above sections have shown that routine practices of court interpreting have the effect of disadvantaging the same lay participants who ostensibly are meant to be served by them: Compared to speakers of the institutional language, they are more likely to be interrupted when their speech is interpreted in consecutive mode, and less likely to understand the entirety of the proceedings, when listening to others’ speech through interpretation in simultaneous mode. They are habitually treated as overhearers rather than as addressees, and they are constrained from drawing on their full linguistic repertoire. These practices have in common that they do not prioritize the communicative needs of speakers of other languages but instead are designed primarily to facilitate institutional processes, and this stays true even if interpreters resist. Reflecting on his own experience interpreting for Tzotzil-speaking Mayan immigrants in the US legal system, Haviland (2019:99) writes that his ‘interpreting engagements, complex as they are, often work not to enhance justice and equality, or even to serve the best interests of the speakers, but rather to lubricate and legitimize the wheels of bureaucracy itself, including its systematic and structural injustices’.

TRANSLATION IN THE LINGUISTIC LANDSCAPE

The notion of translation as a tool in population control is also relevant in linguistic landscape studies, especially for the analysis of signage that is produced by institutional agents for addressees who speak other languages. Kroon, Jie, & Blommaert (2015:4) propose the term public order signs to describe signs that are ‘manufactured by a public authority with the intention of informing the public about an aspect of public order’. They add ‘such signs… often specify what is legally appropriate (and consequently sanctionable in the event of transgression)… they are legally binding both for the authority producing them and for the audiences consuming them’. In linguistic landscape studies, such signs have also been described as directive signs (Mautner 2012), or warning notices and prohibitions (Spolsky & Cooper 1991:76–81), and they fall into the wider category of what is often called top-down signage ‘introduced by governments and big corporations’ (Shohamy & Gorter 2009:3). Jones (2017:153) notes that such signs index surveillance, that is, they imply that ‘someone… is watching’ to make sure that transgressions do not occur.

Multilingual public order signs generally involve processes of translation, as do many other multilingual signs, though this is not often discussed explicitly in linguistic landscape studies. However, the relevance of processes of translation becomes readily apparent where signs include ungrammatical or unidiomatic texts that can be described as translation errors. A striking example is discussed by Barni & Bagna (2010:10), showing a sign from the Italian city of Prato with text in Italian, Chinese, Arabic, and English. This sign is intended to inform the reader that it is forbidden to spit on the ground, based on a communal bylaw that is explicitly noted in small font at the bottom. The English text, distributed over three lines, reads For hygiene reasons/forbidden/to spit on the ground. Both the formatting and the absence of the expletive it is suggest that it is a line-by-line translation of the Italian text at the top that reads Per motivi di igiene/vietato/sputare per terra. Between these two texts, the sign also includes Chinese and Arabic texts. The Arabic text is written backwards, that is from left to right rather than right to left, and with letters not properly connected, rendering it practically unreadable for Arabic readers. This raises questions about how the text might have been produced, but the final production is unlikely to have involved participation from anyone who could read Arabic.

Barni & Bagna (2010:10) interpret this sign as evidence that the local government in Prato recognizes the presence of a community of Chinese migrants. This reflects a common, often implicit tendency in linguistic landscape studies, to view the presence of minority languages as furthering linguistic justice, as a sign of recognition and inclusion, and as a boost to ethnolinguistic vitality (cf. Landry & Bourhis 1997). But as Jaworski & Thurlow (2010:11) have noted, ‘the presence or absence of a language on public signage’ needs to be investigated ‘in combination with the type (or genre) of signs, their contents and style’. If we examine multilingual public order signs as a ‘type (or genre) of signs’, we may find that the inclusion of a language can have a discriminatory effect, by furthering stereotypes. This is because public order signs affirm the existence of deviant behavior at the same time that they reject it, reminiscent of Kulick's (2003:145) analysis of the ‘dual indexicality of no’. The public order sign from Prato in Barni & Bagna (2010) presupposes that some people have been spitting on the ground, or at least are inclined to do so. It also presupposes that the prohibition is relevant to readers of these languages, thereby characterizing them implicitly as people who are likely to require this instruction. Given the fact that the sign displays parallel texts in multiple languages, these presuppositions are accessible to readers of any of the languages involved. So, in other words, Italian readers of this sign may interpret it as evidence that speakers of Chinese, Arabic, or English habitually spit on the ground and need to be told not to do so. Signs such as the one discussed by Barni & Bagna can thus be understood as producing or reinforcing stereotypes about speakers of the languages on the sign. However, this is not how such signs are typically interpreted in linguistic landscape studies. For example, Backhaus (2007:126–28) does not consider this perspective in his discussion of a multilingual warning sign noting the police protection of a bank in Tokyo.⁴ Spolsky & Cooper (1991:89), moreover, discuss the selection of languages on multilingual warning signs in Jerusalem as a reflection of the real-world deviant behavior of their speakers, suggesting that a multilingual warning sign prohibiting ‘immodest dress’ at the Western Wall omits Arabic because ‘Arabs do not dress immodestly’.

Punitive multilingualism

In prior work (Angermeyer 2017), I have proposed the term punitive multilingualism to describe the use of subordinated languages in public order signs. Contrary to a multilingualism that promotes the inclusion of other-language speakers and affirms the legitimacy of their presence, it is a multilingualism of warnings and prohibitions that addresses speakers of subordinated languages as deviant others. This genre is quite common in contexts where institutions are trying to control the behavior of people who are deemed linguistically and culturally different, such as racialized and stigmatized migrants. The discriminatory effect of punitive multilingualism is particularly striking if the respective languages are confined to the public order contexts. This can be illustrated by examining the presence of Turkish in the linguistic landscape in parts of Germany. Figures 1–3 show such multilingual public order signs, part of a collection of similar signs photographed in different German cities and on regional trains.

Figure 1. Warning sign about fines for fare evasion on a regional train in Southwestern Germany.

Figure 2. Signs in a German department store, informing about electronic article surveillance (national chain).

Figure 3. Detail of image in Figure 2.

Figure 1 shows a sign photographed on a regional train in the Rhein-Main area of Germany. It informs the reader in German, English, French, and Turkish about the fine for passengers riding the train without a valid ticket. The text in each language is accompanied by the flag of a country where the language is spoken, drawing the reader's attention to the relevant portion of the text, while also enhancing the recognizability of the languages to viewers. Other multilingual signs found on the same train (not shown here) inform about rules for priority seating and about the functioning of doors and emergency exits. These signs are posted in German, English, French, and Italian. A passenger bill of rights is posted in German only. Turkish speakers are thus singled out in that they are addressed solely as potential fare evaders but are not provided with information that would enhance their safety or comfort while riding the train.

A similar distribution can be found in the corporate linguistic landscape of a department store in Bonn, Germany. Figure 2 shows signs posted at store entrances informing visitors that merchandise is electronically protected against shoplifting (illustrating again the relationship between public order signs and surveillance; cf. Jones 2017). The text is displayed in German in large script on one sign. Below it is a second sign of the same size that is divided into four segments, with translations of the text into Italian, English, French, and Turkish (see Figure 3 for a close-up). As in Figure 1 above, other language texts are accompanied by a relevant national flag. This same sign is found at the entrances of other department stores by the same national chain in other German cities. In contrast to this warning sign, other signage in the store is exclusively in German, except for the store directory, which is provided in English as well. As a result, speakers of French, Italian, and Turkish are addressed only as potential shoplifters, not as potential customers.

The tendency of German institutions and businesses to use Turkish language signage primarily to police deviant behavior has previously been observed by Hinnenkamp (1999), noting also their discriminatory effect. Hinnenkamp also argues that, based on standard language ideology in which the presence of non-official languages is sanctioned only in the absence of understanding of the official language, such signs also index a lack of German proficiency among Turkish speakers, even when this may not be the case. For example, a sign prohibiting ball playing, Ballspielen verboten – top oynamak yasaktir (sic), is ostensibly addressed to ball-playing children. However, such children can be expected to speak and read German.⁵ In view of this point, Hinnenkamp argues that this bilingual sign can actually be understood as encouraging German speakers to monitor the behavior of ‘Turkish kids’ whom the Turkish part of the sign singles out as potentially deviant (Hinnenkamp 1999:16).

Piller (2016:56–59) makes similar observations about the use of Arabic, Chinese, and Vietnamese on signs forbidding smoking or regulating the use of public toilets in Australia, and comparable examples have also been found with Spanish-English bilingual signs in the United States. Notably, Hill (2008:123–24) cites various examples of bilingual public order signs in her well-known study of racist language use in the Southwestern United States, including instructional signs about toilet use such as Wash your hands/Lave sus manos (sic). In a study of the linguistic landscape of a small town in Oregon, Troyer, Cáceda, & Giménez-Eguibar (2015:63) find that the only two uses of Spanish in government-issued signs involve prohibitions and warnings (No alcohol in park/No alcohol en el parque, and Warning – Premises protected by video surveillance/Aviso – Este lugar protegido por vigilencia video). These examples of punitive multilingualism all have in common that they involve languages spoken by groups of migrants who have established significant communities in the respective countries, and who are also frequent targets of racism (e.g. Turkish in Germany, Chinese in Australia or Italy, Spanish in the United States). Taken together, they suggest a relationship between punitive multilingualism and the nationalist language policies of liberal democracies, though the phenomenon is likely not exclusive to such contexts.

Errors and machine translation

Another common feature of punitive multilingualism is that texts in subordinated languages frequently involve ungrammatical or unidiomatic language use, as well as typographic or orthographic errors, such as, for example, the erroneous use of the dotted letter i instead of the dotless ı in Turkish signs in Germany, observed by Hinnenkamp (1999) and also in the sign in Figure 3 where hırsızlığa ‘theft + dat’ is spelled hırsızliğa.⁶ Hill (2008:124) notes that the Spanish translations on bilingual signs in the US are often ‘grossly ungrammatical and unidiomatic, as well as misspelled’, and she observes that this indexes to Spanish speakers that ‘their language is not taken seriously enough to require consultation with them’. By contrast, authors of linguistic landscape studies do not always address such errors, as in the examples from Barni & Bagna (2010) or Backhaus (2007) mentioned above.

Arguably, such errors are not coincidental, but rather a systematic feature of punitive multilingualism, because such signs are regularly produced without the involvement of speakers of the respective languages. One reason for this is in their inherently face-threatening nature, since they presuppose the existence of deviant behavior by speakers of the language. As a result, it is perhaps not surprising if institutions prefer not to ask community members for a translation, in contrast to the use of community member interpreters in face-to-face interaction. Moreover, if sign producers hold racist prejudice towards speakers of the relevant language, they are perhaps generally unwilling to engage personally with members of that community, and the production of a written sign may itself be the result of such attitudes.

In recent years, the availability of machine translation tools such as Google Translate has provided sign authors with a free and convenient method to produce texts in other languages, ostensibly removing the need to consult with speakers or to pay professional translators. Yet texts produced by machine translation frequently involve errors that render them ungrammatical and incomprehensible. For example, in research on public order signs in Toronto that were addressed to Hungarian-speaking Roma migrants, I found that many signs were produced with Google Translate and contained serious errors (Angermeyer 2017). Roma interviewees complained that some of these signs were not comprehensible to them, while others were perceived as impolite. Interviewees felt stereotyped as potential troublemakers and disrespected by the style of the notices. They attributed the ungrammaticality to the use of machine translation, and they took this as indexical of a deliberate decision to avoid face-to-face interaction.

Google Translate and other machine translation applications use a statistical translation method that searches corpora of parallel texts for corresponding strings of source and target words. As a result, the quality of the translation depends greatly on the size of the parallel corpus, which is greater for some language pairs than for others. This unequal development of machine translation resources and the resulting unequal quality of translation are not transparent to users, however, as Google Translate and other providers present the languages in their portfolio as fully parallel and equivalent choices. As a result, the unequal development of machine translation can be said to exacerbate social inequalities between speakers of different languages (see Vieira, O'Hagan, & O'Sullivan 2020:7). In the case of machine translation from English into Hungarian examined in Angermeyer (2017), poor translation quality is manifested by a general absence of case marking on nouns and of person or tense marking on verbs, which is a major source of the ungrammaticality in these ostensibly Hungarian texts. Another source of problems arises from the fact that machine translation translates series of text fragments without creating a coherent whole, as noted by Pym (2011:4). This reflects again the ideology of referential transparency that locates the task for translation solely in matching words from one language to words in another language. In multilingual signage, such textual fragmentation is often evident when typographic line breaks are treated as syntactic boundaries (see Angermeyer 2017:173), a phenomenon that is also apparent in the translation from Italian to English on the sign discussed by Barni & Bagna (2010, and see above).

Despite such shortcomings, users of machine translation seem to have remarkable faith in the technology and downplay its limitations (see Angermeyer 2017:175). This includes institutions such as the Toronto city government and its public school system, which use Google Translate extensively on their public webpages and advertise this as furthering the accessibility of public services by speakers of other languages, in a city where half the population are immigrants.⁷ Machine translation thus creates new opportunities for linguistic injustice and abuse, particularly in the context of public order signs. It is easy to imagine a situation where the text on a prohibition sign is entirely incomprehensible, but addressees are nonetheless sanctioned for failing to follow the intended directive. Abuse of machine translation also exists with translation into institutional languages, particularly in the context of surveillance, as illustrated by an incident in 2017, when a Palestinian man was arrested by Israeli police after Facebook's machine translation tool rendered his Arabic post meaning ‘good morning’ with a phrase meaning ‘harm them’ in Hebrew (Lau 2021:9). As in the case of erroneous translation in punitive multilingualism, this case shows an institution putting blind faith in technology, instead of seeking verification from speakers of the subordinated source language, who are distrusted and suspected of deviant behavior. As noted by Heller & McElhinny (2017:181–88), government funding for the development of machine translation helped fuel the growth of linguistics departments in the US after World War II. Arguably this history gives sociolinguists some degree of obligation to pay close attention to the actual use of machine translation and its social consequence, though there have not been many studies of this to date (see Jacquemet 2019).

CONCLUSION

In conclusion, this article has shown that translation is an important topic for sociolinguistic research that should not be left entirely to scholars in the field of translation studies, particularly if we understand it as part of a range of communicative practices in multilingual contexts, alongside phenomena such as codeswitching, translanguaging, and second language use (see e.g. Tipton 2019; Maryns, Angermeyer, & Van Herreweghe 2021). Translation can also be investigated in relation to processes of language contact, language shift, language differentiation, and language standardization, all of which may be influenced in part by how translation is done, by whom, and for what end. This article has shown that linguistic injustice can arise from practices of translation and interpreting that disadvantage and disrespect speakers of subordinated languages. This occurs, for example, when translation is used asymmetrically, that is, when opportunities to speak without interruption and to check one's understanding are not distributed equitably. Participants are also disadvantaged by the use of a formal, institutional register that rejects their own vernacular language use, or by the use of careless, ungrammatical forms that may not be comprehensible and that may index negative stereotypes about them. All of these issues have in common that translation is not geared to accommodate to the speakers of the subordinated language, but instead aims to ensure that institutional practices are not disrupted by linguistic diversity. In other words, we could say, paraphrasing Rosa & Flores (2017:642) that interpreting in court and punitive multilingualism in the linguistic landscape actually serve as mechanisms ‘for producing governable subjects’ in support of ‘the raciolinguistic status quo’. This happens despite appearances to the contrary. As Hinnenkamp (1999:16) notes, translation may give the impression of tolerant pluralism and equitable information, masking the discriminatory effect. This parallels Hill's (1998:683) observation that Mock Spanish can index a ‘cosmopolitan’, ‘congenial persona’ at the same time as it functions as ‘covert racist discourse’ whose understanding requires ‘access to very negative racializing representations of Chicanos and Latinos’. This suggests that unjust translation practices and linguistic injustice are as much a symptom, as they are a cause of injustice. Yet by striving to overcome linguistic injustice and advocating for more equitable translation practices perhaps we can help facilitate resistance towards other forms of injustice, resisting both the underlying language ideologies and the ideologies about race, nationalism, and state institutions to which they are tied.

More just translation in institutional settings could begin by respecting speakers of subordinated languages as interlocutors and by making sincere efforts to check their comprehension. In fact, the above-mentioned guidelines for the communication of rights of suspects in police custody point in this direction, as the authors argue for a legal standard of ‘demonstrated understanding’. This is to be demonstrated not by asking yes/no questions such as ‘do you understand?’ but rather by asking individuals to restate these rights in their own words (Communication of Rights Group 2016:5). In line with these guidelines, sociolinguists more generally might want to put more emphasis on the study of comprehension and understanding, and advocate for understanding between people, rather than translation between languages.