A voice in the crowd: Broader implications for crowdsourcing translation during crisis

1. Introduction

Using crowdsourcing for translation in the context of a crisis, such as a natural disaster or violent political upheaval, is different from using crowdsourced translation for other uses for two reasons. First, mistakes made in translation during a crisis could result in the loss of human life. Second, but not as immediately apparent, the translated information generated from these crowdsourced applications becomes the basis for policy-making about the crises themselves. In this paper, I concentrate on three areas of weakness in the design process: flow of information, expertise of translators, and loss of contextual markers. After explaining why each element is significant for developing applications used in a crisis context in general terms, I will illustrate the development process failures in four crises: post-earthquake Haiti, Arab Spring Egypt and Libya, and more recently Somalia. By understanding the consequences of the failure, the intention of this paper is to provoke a shift in the design process. This is not a criticism of one particular organization or any specific crowdsourcing application, but rather the overall approach to design which does not incorporate the distinguishing features of the crises, the real-life stakes for the individuals being translated or the expertise of translators. Observations about the approach to design come from my practitioner experience in conflict resolution and specific examples are based on my work as a volunteer translator for 4636 Haiti in 2010 (a crowdsourcing translation project) and as a political scientist in North Africa during 2011.

Within the current system of crowdsourcing translation, the sources who provide the information through their voices, their messages, are absent from subsequent decisions made with that information by international actors – crucial decisions, such as in the scramble to organize international aid in post-earthquake Haiti. Changing the design process so that sources were included in the decision-making chain, cultural communication experts were consulted, and contextual information was retained would deepen applications’ impact for crisis victims.

In each situation, Haiti, Egypt, Libya and Somalia, calculations about the future of a nation were being made. Development contracts were being negotiated by foreign actors. Policies were being written by international agencies. The voices of the people on the ground in those countries were available to decision-makers in a way previously unimagined because of crowdsourcing projects; however, the failure to design translation crowdsourcing applications for a crisis context resulted in these victims of crisis being unable to access the information they contributed while simultaneously being bound by that translated information as it shaped their political environment from the outside.

In this section, the impact of not considering the crowdsourcing design process at the conceptual level will be discussed in general terms. Later, in Sections 2–5, specific examples will illustrate how the information collected from crisis victims was used to solve the crisis and perhaps plan for an entire nation.

1.1. Structure: analysis of process flow

The traditional definition of crowdsourcing focuses on the crowd, the group working on the task presented by the crowdsourcer (Figure 1). As M.N. Wexler describer the relationship:

The crowd is depicted as a problem solver and innovator … [and] the crowdsourcer seeks to extract the benefits and privileges from the crowd. The motive is not that of a co-creator but an instrumental user of the value to be conferred by calling, filtering and managing the crowd. [1]

OPEN IN VIEWER

Figure 1.

Information management of the crowdsourcing process [1].

Crowdsourcing was conceived as a business solution, the next technological evolution beyond outsourcing, but its use for translation during a crisis mandates a change in design approach. Howe’s often cited essay in Wired [2] gives many early examples of how crowdsourcing has been used. In each instance, the crowdsourcer, whether a business or an organization, planned to accomplish a large task by harnessing the labour of the crowd. Howe described a pharmaceutical company’s research and development that was pursued through crowdsourced initiatives run like contests, the birth of iStock photo, which capitalized on the surplus of crowd-generated images rather than professional photographers’ work, and Wikipedia.org, which marshalled talented coders to build an enormous encyclopaedia edited and maintained by the crowd. In each instance, the crowd is manipulating or contributing information directly related to the task. For example, on Wikipedia, a crowd-member edits information on the Wikipedia site for other Wikipedia users. Also, the product of the task is explicitly defined. For example, in research and development, the crowd is called together by a business or organization to generate a concrete product. The individual from the crowd who generated the idea for the product may continue to participate in the decision-making process related to that idea. It is a linear process flowing from input to task to product along which the organizer of the crowd and members of the crowd have a role and a mode of participation.

Translation and crisis are two variables that dramatically change this process in two ways which can be seen in the ‘structure’ of the process. First, there are in essence two inputs and two crowds. Developers have not distinguished between the source and input or the victims and crowd. Unlike an application designed purely for translation, such as the machine translation application Google Translate [3], during crisis situations, translation is an intermediate rather than an end task. For this reason, the confusion starts with the term input. It is used to describe both the source data and the translated information completed by the crowd used for the next stages in the process (next stages include disaster relief and policy planning to promote disaster relief). For example, as a translator after the 2010 earthquake in Haiti, I read SMS messages in Kreyol (the source information) that I translated to English. The English version became input to be aggregated and available for statistical manipulation. It became the reservoir of information for disaster relief organizations, inter-governmental agencies and the news media. There was a crowd of disaster victims contributing information, and a crowd organized around the task of translation. The design process for crowdsourcing cannot adequately incorporate two crowds and two inputs. Without a distinct term, the sources seem to drop away and are forgotten by developer teams, which include software engineers, crisis management experts and political policy analysts. To use the example of Google Translate again, this application invites participation from the crowd, one crowd; its design can only accommodate one notion of crowd. That is perhaps why in crisis applications only one crowd, the translators, can participate in accessing the information. The sources of the information, the crisis victims, do not have a mode of participation beyond their first contribution. They have no role in the process after the first step. It is the second crowd that becomes the focus of the design process. Why does this matter?

That is the second part of the problem. The beneficiary during a crisis is not the crowdsourcer as Wexler described in the business solution. In a crisis, the beneficiary is the same group of people contributing input for translation; it is the victims of the crisis. However, the design of the crowdsourcing application disenfranchises the true beneficiaries from the product, the information. The crisis victims should have the same opportunity to make decisions with the information they contribute as outside actors (such as international relief organizations or governmental agencies) have with the translated version. To take an example with socio-political impact, citizen journalism also leverages the submissions of the crowd. The input such as videos and photos is crowd-generated [4]. The crowd that participates in the input process and often the filtration and management process retains some control over the destination of the input they contributed [5]. However, in translation crowdsourcing, the crowd is an intermediary group, not the group that generated the material. After the original material crosses the language barrier, the sources are cut off from it by the current crowdsourcing process. The second crowd, task and information management continue in a new language, thereby cutting out the crisis victims who stand to benefit from the information. Without translation working in the opposite direction, the original sources may not be able to access the information they contributed or participate in determining its destination.

If there were to be a diagram to illustrate the absent sources, it might contain a circle of people representing the disenfranchised crisis victims exterior to the linear sequence of the rest of the crowd/task/information flow. Ushahidi, an open source crowdsourcing platform for information management including products for crowdmapping and translation, created Swift River, which combines real-time information filtering and translation. Figure 2, created by Ushahidi, illustrates the relationship of the source, crowd and application filtering or managing the information flow.

OPEN IN VIEWER

Figure 2.

Diagram depicting flow of information in Swift River relative to sources [6].

In this instance the crowd is the filter along with several digital tools. It was not at all intended to illustrate the point of removing the source from the sequence; however, it is interesting to note the way in which another individual, who is a member of the crowdsourcing design community, visualized the system by moving the human sources exterior to the path of information flow. Another way of interpreting the diagram is to notice how it captures the confusion of the two crowds and also fails to differentiate between the two inputs. The design process has not adapted to the unique circumstances of translation during crisis. The applications designed from this process do not address the needs of the sources who are the real beneficiaries of this tool. Examples from projects in Haiti, Libya and Somalia will further illustrate the impact of this structural flaw in action.

2. The case of Haiti 2010

The humanitarian efforts in post-earthquake Haiti illustrate the structural flaw and the failure to include cultural communication experts in the design process. After the earthquake on 12 January 2010, an emergency SMS number was set-up for victims to broadcast their locations and urgent needs in a collaborative effort between the region’s major cell phone carrier Digicel, humanitarian aid organizations and engineering teams (Crowdflower, Samasource, Instedd, FrontlineSMS:Medic, Google.org, MIT Media Lab, Sahana, Stanford University, US State Department, Ushahidi and Votident). 4636 Haiti aggregated emergency text messages, and with a team of global volunteers, translated messages from Kreyol and French to English [20, 21]. Key information necessary to dispatch a rescue team on the ground, such as the International Red Cross, was extracted by the translators and entered into an online form. Hours after the earthquake, this form went into development miles away from the crisis by a team that was expert in collecting and moving information in consultation with a team that was expert in disaster response. From my experience, the application was not designed in consultation with Haitian policy experts or Kreyol speakers. Disaster victims were not treated as a separate crowd and they were not treated as the end beneficiaries of the task. In a 2010 interview with the tech magazine Venturebeat, one of the co-creators of the project, John Nesbit, remarked:

Honestly, this is rare to see groups like the State Department, Ushahidi and Instedd all working together … I hope it doesn’t take another catastrophe to see this type of collaboration again. The bright spot in all of this is seeing the tech community take ownership. [21]

His interviewer Cutler added:

True, but the more intriguing part of the story may be that this all started with a simple tweet. In fact, Nesbit never set foot in Haiti. [21]

The interviewer pointed out the important fact that the main developers had no experience with the context of their design. They had proceeded in the design process as though this were a traditional crowdsourcing project. By not inviting cultural communication experts or Haitian policy experts, the development team remained isolated from the unique circumstances of crisis and translation that might have caused them to alter their design process. The following examples of SMS messages from Haitians just after the 2010 earthquake demonstrate the shortcomings of a system designed for translation without the consultation of cultural communication experts familiar with the language data about to be managed. While the goal of the project was to coordinate relief efforts and reach victims, it did so through a translation crowdsourcing application that promoted the aid organizations as the end beneficiaries. The application was intended to act like an emergency service number in the United States or Europe. If that was indeed the model, then victims would have been able to access statistics about those services, access the information they contributed, and make decisions about their community with that information. The application designed for Haiti did not allow earthquake victims to participate in this way. The application was not conceived with the victims as the end beneficiaries of the information. The application’s online interface had a space for: name, age, gender, location, nature of emergency, etc., which generated a form guiding disaster relief workers to victims.

SMS 1: In the earthquake I lost four people not counting an amputated foot.

SMS 2: Name: Br – profession: dock worker: telephone number – I am asking you to find me a job because my house was cr[ushed]

SMS 3: What role will teachers play in the reconstruction of Haiti? [22]

However, as illustrated by the examples above, many of the SMS messages comprised information that would not fit into such a simple form. When translators encountered messages like these, they were instructed to mark a box titled ‘not enough information’, indicating that there was no need to send an emergency response team. The information was archived; however, the access of this archiving structure was in English. If a Kreyol speaker wanted to compile a list of all the SMSs mentioning education, as in SMS 3, these messages would be nearly impossible to retrieve. As the aid money was promised, this application housed information about how Haitians wanted the money spent. Owing to the language barrier in the application, the invisibility of this initial crowd of sources, failure to consider the disaster victims as end beneficiaries and the failure to include culture communication experts in the design process, Haitians were unable to leverage this information. The architecture of this application in Haiti was designed to coordinate relief efforts during a humanitarian crisis, not to conduct a poll or provide political transparency, but there is no reason not to design with all these aims. Implementing an application during a major disaster or conflict must entail an examination of the complex environment where that application will be used. While Haiti may not have a tradition of democracy and open information that is connected with the model of having access to information from an emergency service call, why should the applications developed preclude that possibility? Designing for a crisis context should require thinking about the broader social and political context of use for the application beyond the immediate task of translation.

Illustrating the institutionalization of the structural flaws in the design process, a UN Foundation sponsored report, authored by the Harvard Humanitarian Initiative, characterized how organizations view the challenge of developing an application that serves the sources, crowd and beneficiary. The report conceded:

[Humanitarian organizations] are just beginning to understand what it means to have two-way conversations with affected populations. ‘Beneficiaries now have a voice, and affected populations have a voice. They’re not just recipients, […] they have the ability to talk back. That [two-way communication] creates a different dynamic around accountability and responsiveness. It also creates a new set of ethical responsibilities, especially around expectations and whether they can be met. […] [Humanitarian] organizations have always prided themselves with being responsive to beneficiaries’ needs, and being accountable to them. But there is a now different set of tools to make that happen and it is taking some organizations by surprise’ – Katrin Verclas, MobileActive. [23]

From the perspective of humanitarian agencies who participate in and inform the design process of translation crowdsourcing applications, a re-conceptualization of the process is just beginning.

3. The case of Egypt

As the revolutions spread across North Africa and the Middle East (MENA), the desire to engage via social media seemed to intensify among participants and observers of the events. The saturation of images of banners in Arabic and videos of protesters speaking in Arabic demanded a new translation strategy for those outside the region to stay current with the flow of information. The cultural divide was evident, so context became crucial to grasp the situation. This section will examine two projects involving Twitter and their treatment of contextual content. In each project, volunteers were called to participate in translation crowdsourcing. The first application was a collaboration between Twitter and Google called Speak to Tweet that circumvented the problem of internet connectivity by allowing cell phone users to call their tweet into a voicemail to be transcribed and translated [7]. Twitter used a crowd of volunteers to enhance translation of Egyptian Arabic after the Tweets were first transcribed by a Mechanical Turk application trained from an initial 10 hours of speech [24]. The company had been using volunteers in this manner since 2009 to broaden its language services [25]. This crowdsourcing application emerged as a leader in bridging the language barrier and allowing information to reach multiple language groups. Lawerence Lessig, quoting Mark Stefik’s, The Internet Edge: Social, Technical, and Legal Challenges for a NetworkedWorld, warned there is a risk in choosing one technology, relying on one digital code. This is particularly relevant in the world of humanitarian aid and conflict resolution in which the same group of actors is always engaged. In addition, among these actors (e.g. UN or partners), there are only a few that have developed crowdsourcing translation applications. Those that do, such as Ushahidi-based products, are quickly adopted by most other actors. Lessig stresses the power of code which:

determines which people can access which digital objects. How such programming regulates human interactions … depends on the choices made.

Lessig continues by summarizing the ethical implications:

a code of cyberspace, defining the freedoms and controls of cyberspace, will be built. About that there can be no debate. But by whom, and with what values? [26]

Odejobi and Adegbola [27] concur that the foundations of translation technology architecture do not incorporate the cultural context. From the Speak to Tweet translations, there emerged one homogenized well of information from which the international audience could drink.

The non-profit Meedan.org was launched in 2005 as a cross-cultural communication space online, in particular between English and Arabic. Meedan developed a project focused on translation of tweets in 2011. Initially capitalizing on the Speak to Tweet material, it sought to provide a way for Arabic and English twitter streams to connect and interact unimpeded by a language barrier. The project used the platform curated.by to integrate the Twitter streams in two languages.

The Meedan curated.by feeds allow readers of both languages to follow tweets originally written in Arabic and English, allowing fuller access to news and commentary emerging live, whether on protests in Syria or presentations on social media and nonprofits at South by Southwest. [8]

In addition to translation, the crowd was invited to ‘curate’ the space with additional information that might add context to the translated material. Meedan gave 10 suggestions for ‘collaborating’ in the curatorial process. A request by the author for comment and clarification about methodology and administration of the site went unanswered by the organization as of 12 January 2012. What was the origin and purpose of the 10 suggestions for curating the dual twitter stream? For example:

#6 Is the Twitter Stream openly political? What does the profile say?

#9 What content is this Twitter Stream reporting? Is it strongly opinionated or dispassionate and factual? [28]

Despite the option to switch the site’s language from English to Arabic, the curating instruction page did not appear to exist in Arabic [28] and the associated programme description page reverts to English [8]. Were contributors encouraged to provide this information about their own streams or deduce it about others? Were translators involved in building the application for translation and the curatorial process? Who can access the information? How was it archived and aggregated? While developed as an intercultural platform, there remained a disconnect between the technical, translation and source communities as presented through the semi-bilingual website Meedan. The aim of the project to contextualize the flood of information was rooted in the understanding that there was a significant cultural divide between the actors in the MENA region and the audience in Europe and the United States; however, who was providing the context? Was there a feedback mechanism to correct misconceptions? Were local MENA actors consulted in the design or just bilingual speakers? While these questions border on other research outside the scope of this paper such as participant reliability, the difference here is in examining the initial design of the application that frames how participants contribute.

The call for volunteer translators provided a solution that software could not. While this crowd served an important purpose, to unblock the flow of information from one language to another, they were amateurs and the platform was developed from a crushing need to relieve this block without the time for critical reflection about where the information would flow. If information is power, there were many actors interested in the information streaming from the region. Once the messages crossed the language barrier, the sources of that information lost the ability to participate in decisions made with that information, just as the earthquake victims in Haiti had. Imagine a protester in Tahrir Square holding up a banner in Arabic. These words were translated, aggregated, warehoused. If that protester wanted to determine how the banner was translated, if the double entendre referencing an ancient poet’s rebuke of military police was appreciated [29], this would be impossible. If political decision-makers were perceived to be using that information in opposition to its intention, there was no feedback mechanism within the crowdsourcing application. Why does this matter? These cultural, contextual references are the foundations of collective consciousness, of societal identity. They may indicate values and priorities that have relevance when considering how to build trusted and cohesive policies about government [30]. If translation tools are intended to support a group coming together to transform their nation, it is vital these tools do not impede their ability to convey their identity in this action by jettisoning contextual markers [31].

In response to MENA’s upheaval, the US State Department launched its own Twitter feed in Modern Standard Arabic, which is powered by Google Translate. One of the initial tweets relayed:

The US foreign ministry has recognized the historic role of social media in the Arab world and we want to be part of your conversations. [32]

This conversation relies heavily on Google Translate, which does not support local dialects. While this formal Arabic is used on Al Jazeera and may be appropriate as government speech [32], the formality is both unusual for Twitter and incompatible with the intent of reaching the masses since the variation between dialects across the Arab world can be as much as that of closely related languages such as Spanish and Italian. This was intended to initiate two-way communication, and to address the flaw perpetuated by other translation applications that discarded the sources; however, it has fallen short of achieving this goal.

4. The case of Libya

Preliminary analysis of one project in Libya points to structural issues as well as lack of consultation with language experts impacting the efficacy of translation crowdsourcing. NATO’s involvement in Libya’s revolution comes with its continued support post-Qaddafi. During a 17 June 2011 panel discussion podcast from the US Institute of Peace, Patrick Meier, director of Crisis Mapping Ushahidi, shared that the US State Department was interested in partnering with his organization again (as it had after the earthquake in Haiti) to conduct a needs assessment of the country. The potential of crowdsourcing information from Libyans filtered through an unspecified format was viewed as more efficient and accurate than the traditional methods of the US State Department [33]. If their previous pairing in Haiti is predictive of their approach, an unchanged formula would mean failure to consult with cultural communication experts, and the disenfranchisement of the original sources not viewed as end beneficiaries. The UN Office for the Coordination of Humanitarian Affairs (OCHA) contracted with Meier to create a crisis map of Libya built from on-the-ground reports filtered through the volunteers of the Standby Taskforce [34]. The After Action Report commissioned by the UN describes political decision-makers and sources, but the sources’ access to decision-making and to the product ‘Libya Crisis Map’ was not considered in the discussion of best practices OCHA [35]. During a panel discussion at the US Institute of Peace, Roohafza Ludin, a graduate of the Ushahidi school [36] from Afghanistan, expressed a reservation about the lack of contact between the source and the final destination of the information. Summarizing a challenge in crowdsourcing related to the enormity of information and the disconnect posed by acting on aggregated data, she concluded:

sometimes you need the voices of the people coming from them directly. [33]

This comment underscores the importance of translators because they provide the bridge to a disconnected, yet indispensable part of the system, the human sources. It also an acknowledgement of the structural flaw, the discarding of the sources, from within the development community.

5. The case of Somalia

In December 2011, the English language Al Jazeera launched an SMS polling initiative with Ushahidi designed to give a voice to the people of Somali and share a picture of how current violence is impacting everyday lives. A call for translators in the diaspora, particularly Somali student groups, was issued online, and phones were distributed on the ground throughout Somalia so multiple users could participate. In terms of design, Ushahidi recycled the plugin developed for Haiti, although the interface has fewer elements. Translators perform the translation task as well as marking a box that categorizes and aggregates the information (see Figure 3).

OPEN IN VIEWER

Figure 3.

Screenshot of interface seen by translators [37].

From these aggregated categories, a map is created to visualize the nature of the crisis. The categories are colour-coded on the map. Instructions to the translators permit additional categories to be suggested, but so far, none have been added to the map visualization or otherwise incorporated into the analysis (see Figure 4 [38]). The stated goal of the project is to give a voice to the Somalia people, but the Somalis who participate have no say in how their voices are categorized or depicted on the map. The developer team and volunteer translators created the categories and the visualization from the crowdsourced translations that frame the discussion. The SMS poll asks an open question:

How has the Somalia conflict affected your life? [39]

OPEN IN VIEWER

Figure 4.

Map visualization of aggregated SMS messages [36].

The respondents are not aware that their replies will be sorted into categories. In one response example:

The Bosaso Market fire has affected me. It happened on Saturday. [40]

The response was categorized as ‘social’. Why does the fact that violence happened in a market, an economic centre, not denote ‘economic’ categorization? There is no guidance for maintaining consistency among the crowd, nor any indication of how the information will be used later. It is these categories chosen by the translators, these bright colourful circles on the map (see Figure 4) which are speaking to the world, not the Somalis. Their voices have been lost through a crowdsourcing application that was designed with a language barrier. They cannot suggest another category that better suits the intentions of their responses. The danger is that these categories become the framework for aid donations and policy endeavours; the application frames the discussion rather than the words of the Somalis. The simplistic categories become the point of departure for aid agencies and policy-makers to understand and become involved with translated material.

Al Jazeera intends a second phase in which they will continue the dialogue with some respondents creating a type of feedback loop. It is unclear whether crowdsourcing will continue to play a role in this stage, or whether it will be purely journalistic. This interaction would not correct the process issues with translation crowdsourcing, such as loss of authorship and disenfranchisement, because the respondents would still not be able to access the information nor manage how it is aggregated and presented to the world. An 8 December 2011 comment on the Ushahidi blog described the most compelling reason why translation in crisis situations cannot be treated like other tasks performed by crowdsourcing, and Ushahidi director Patrick Meier, responded to the comment:

A----, My friend received the message from you on his phone. The question says ‘tell us how is conflict affecting your life’ and ‘include your name of location’. You did not tell him that his name will be told to the world. People in Somalia understand that sms is between just two people. Many people do not even understand the internet. The warlords have money and many contacts. They understand the internet. They will look at this and they will look at who is complaining. Can you protect them? I think this project is not for the people of Somalia. It is for the media like Al Jazeera and Ushahidi. You are not from here. You are not helping. It is better that you stay out.

Patrick: Dear A----, I completely share your concern and already mentioned this exact issue to Al Jazeera a few hours ago. I’m sure they’ll fix the issue as soon as they get my message. Note that the question that was sent out does *not* request people to share their names, only the name of their general location. Al Jazeera is careful to map the general location and *not* the exact location. Finally, Al Jazeera has full editorial control over this project, not Ushahidi. [38]

As of 14 January 2012, there were still names featured on the Al Jazeera English website. Further analysis of the impact of this collaboration is not within the scope of this paper. It serves as an example of the structural flaws in the crowdsourcing process. The consensus among the engineering community has been that crowdsourcing is beneficial. Anastasiou and Gupta describe the potential for crowdsourcing translation in optimistic terms:

Apart from the enterprises, the community gains from crowdsourcing in that it can use the technology in its own mother tongue, and thus its voice can be easily heard – which is particularly important for minor languages … in crowdsourcing initiatives we witness reinvestment of translation technology to further human concerns or agendas. [16]

However, there is evidence that there remain obstacles to being ‘easily heard’. When crowdsourcing removes authorship, dissolves intentionality, context or other cultural elements, it is not an effective tool for transmitting communication information. Decisions and policies developed from the translated information are consequently less connected to ‘real voices’ than decision-makers at the final end of the information chain believe.

6. Conclusions

When a source cedes control of his/her voice by providing information to be translated, s/he says in essence, I want to be heard and understood. As part of the team that develops applications for translation, we are not simply responsible for the faithful conveyance of meaning and intent of that source’s voice, but we tacitly agree with the form in which that voice becomes archived, becomes input. The current approach to design denies the sources a mode of participation with that information once it crosses the language barrier. Without the ability to capitalize on the information they provide, that information supports the decisions of foreign actors such as international non-governmental organizations or intergovernmental agencies who are traditionally seen as the end beneficiaries of the crowdsourcing applications rather than the crisis victims themselves. The unique circumstances of translation and crisis demand an approach to design that can differentiate between more than one input and crowd, will consider the crisis victims to be the end beneficiaries, consults with cultural communication experts and retains contextual information. The traditional or business approach to design focuses mainly on achieving the aims of the task, but in a humanitarian disaster or political conflict, that is myopic and irresponsible. We have reached a point at which we have several applications that show promise and have been field tested. We can reflect on best practices and not feel the pressure of urgent response to a critical situation. Best practices include ethical considerations and big picture thinking. What role will the application play in the information ecosystem of the crisis? Will the application collect information that could be useful to political decision-makers? Who can access information? Can we build an application that can perform the immediate task of translation and long-term task of bi-lingual information management? The examples from Haiti, Egypt, Libya and Somalia provide instances of applications and projects which neglected to measure their impact beyond quantity of translations. Each example illustrated how the approach to design failed to incorporate the unique circumstances of the situation so the resulting applications restricted rather than empowered crisis-affected populations.

The speed of development of crowdsourcing translation and the volume of information it can convey should not be the excuses for our failure to reflect and research, to distinguish between the medium and the content. The qualities of translation crowdsourcing in crisis dictate the importance of understanding much more about its potential and future global impact.