How AI Translation Compromises Asylum Applications

In recent years, the UK has seen a growing backlog of asylum seekers, creating a significant administrative challenge for the government. Currently, around 118,329 people are waiting for decisions on their asylum claims.

This situation creates a bleak reality for many asylum seekers, who are prohibited from working, have no say in where they live, and must rely on government support of just £7 per day to cover essentials like food, sanitation, and clothing.

A significant number of asylum seekers—16,031 by early 2024—have also been placed in detention centres, where they often endure harsh and difficult conditions. It is clear that there is a pressing need to address this issue.

AI’s Potential Role in Addressing Asylum Seeker Backlog

Recognising the urgency of this issue, the new Labour government has committed to prioritising the reduction of the asylum backlog. In a recent speech, they promised to bring relief to those who have been trapped in the system for years, including many who have been forced to live in hotels with limited opportunities to work. This decision has been widely welcomed, as it brings hope that many people stuck in the asylum system may finally receive a grant of status and be able to move forward with their lives.

In an attempt to address this issue previously, the government introduced a controversial method: replacing in-person interviews (with interpreters) with a 50-question questionnaire, available only in English. Asylum seekers were advised to use Google Translate if needed, a suggestion that has sparked much criticism.

While this approach may speed up case processing times, it risks depriving asylum seekers of the opportunity to fully explain their circumstances, especially those who find it difficult to articulate complex experiences in written form.

Given the rapid rise of artificial intelligence, it’s no surprise that many countries have begun exploring AI as a solution to the growing asylum backlog. In line with this trend, the UK government has recently launched a ‘short inquiry’ into,

“The current capability and accuracy of market leading artificial intelligence and machine translation tools in relation to ITS (Interpreting and Translation Services in courts).'”

This inquiry suggests that the UK government may be open to leveraging AI in the future to streamline the asylum process and reduce the backlog.

Risks of AI Translation for Asylum Seekers

The use of AI translation tools in such high-stakes situations introduces significant risks, particularly when dealing with the delicate and often traumatic personal narratives involved in asylum claims.

Although fast and efficient, machine translation tends to miss the critical cultural and contextual nuances that human interpreters are trained to recognise. Asylum seekers often recount experiences of persecution, violence, or complex legal and socio-political challenges in their home countries. These narratives are deeply personal and can involve layers of meaning that automated systems may fail to capture, especially when the language involves idiomatic expressions, emotional undertones, or references to specific cultural practices.

A mistranslation or misinterpretation of key details can mean the difference between an acceptance or rejection of a claim, potentially placing individuals in harm’s way, facing deportation back to dangerous environments.

In addition to these general concerns, AI translation tools face particular challenges when translating ‘low resource’ or minority languages. AI models rely heavily on large datasets to generate accurate translations, but many of the languages spoken by asylum seekers—especially those from regions like sub-Saharan Africa, the Middle East, and parts of Asia—are considered low resource in AI. This means there is insufficient data available for the AI to “learn” these languages properly, leading to translations riddled with errors, omissions, or outright “hallucinations”.

These risks disproportionately impact asylum seekers who speak minority languages, further widening the gap in their ability to present their cases effectively. Without access to human interpreters who can fully grasp both the language and the cultural context, asylum seekers from marginalised linguistic communities may find themselves at an even greater disadvantage, exacerbating the already unequal power dynamics in the asylum process.

Lessons from the US on The Dangers of AI Translation

We have already seen the pitfalls of relying on AI translation tools in the United States, where similar attempts to streamline asylum processes using technology have encountered serious issues.

In particular, the US courts have used AI translation to process asylum claims from Afghan refugees, but the outcomes have been catastrophic for some asylum claims. The implementation of neural machine translation tools—designed to automate and expedite the translation of personal statements—has resulted in a number of translation errors that have significantly undermined the integrity of these cases.

One of the most critical problems identified is the misuse of personal pronouns in translated statements. For example, when asylum seekers recount their personal stories of persecution or escape, these AI-driven tools mistakenly swap “I” for “we” or vice versa. These seemingly small errors can introduce serious inconsistencies in the applicant’s account, raising doubts about the credibility of their testimony in the eyes of immigration officials and judges.

In asylum cases, where the stakes are incredibly high, even the smallest perceived discrepancy can lead to a rejection. These inconsistencies often cast asylum seekers as unreliable witnesses of their own trauma, making it harder for them to prove their need for protection.

The consequences of such errors are severe and far-reaching. In many instances, asylum claims have been denied outright due to these inaccuracies, effectively ending the refugee’s chance of gaining protection and legal status in the US These are cases where human oversight could have easily caught the translation errors and clarified the individual’s narrative, ensuring that justice was not derailed by a simple language mistake.

Unfortunately, by removing human translators from the equation and depending too heavily on AI, asylum seekers are left vulnerable to being misunderstood, misrepresented, and ultimately denied the refuge they so desperately need.

The Need for Urgent Human Oversight in AI Translation

This experience from the US serves as a cautionary tale for other countries, including the UK, which are exploring similar AI-driven approaches. It highlights the critical need for human oversight in the asylum process, especially when dealing with sensitive personal narratives that require not just linguistic accuracy but also a deep understanding of cultural context and individual circumstances.

While powerful, AI lacks the emotional intelligence, cultural awareness and data from low-resource or minority languages that human translators bring to such high-stakes scenarios. Without these, the risk of failing the people who most need protection becomes too great.

At AST Language Services, we’ve always emphasised the importance of combining human expertise with advanced technology. Though we use sophisticated tools like translation memory and CAT software, we make sure that human translators are central to the process. This guarantees that the subtle cultural and emotional aspects of language are captured, avoiding the critical errors that using AI alone might introduce.

When it comes to matters of life and safety—such as asylum seeker claims—AI can be helpful, but human oversight is absolutely irreplaceable.