Navigating the Language Barrier with Image-Based Machine Translation
For newly arrived international students in Finland, the language barrier is often one of the first and most persistent challenges they face. While professional translation is not always accessible or feasible in everyday contexts, free online machine translation (MT) tools offer a practical alternative. Among them, Google Translate stands out as one of the most widely used, particularly due to its integration with Google Lens, which is a feature that enables instant image-based translation without requiring users to type or copy any text manually.
For my master’s thesis, I carried out a pilot study examining how Turkish-speaking students in Finland engage with this camera-based MT technology. Specifically, it explored not only how they use the tool but also how they perceive and evaluate its output in practical, real-world contexts.
Five students, all newly arrived and unfamiliar with Finnish, were asked to translate four types of everyday printed texts using Google Translate’s camera function: a recipe, a local news article about bear sightings, a speeding ticket, and a medication pamphlet. The varying text types were chosen to evaluate the tool’s capabilities in translating different contexts, layouts and designs. Afterward, the students reflected on their experiences in semi-structured interviews.
Evaluation of Perceived Trustworthiness
“It works fast, but you cannot really trust it blindly,” one student noted, a sentiment shared by all participants as each session revealed recurring issues with mistranslations, awkward phrasing, and varying degrees of error across each text.
When asked to rate their trust and confidence in each translation, the following average scores were recorded:
- 78% for the medication pamphlet
- 74% for the recipe
- 37% for the news article
- 23% for the speeding ticket
These ratings were influenced not only by the number and type of translation errors but also by the perceived stakes of each scenario. For instance, while a minor mistake in a recipe might be deemed inconsequential, misunderstandings related to legal or medical content raised greater concern among the students.
Importantly, not all issues were linguistic in nature. Optical character recognition (OCR) errors were also common. In one case, a speeding ticket was misread as a string of Arabic characters, confusing the participant who then decided to skim through the translation before retaking the photograph for a second attempt.
Adaptive Strategies and MT Literacy
Despite disruptions, participants rarely gave up on the task. Instead, they demonstrated a range of adaptive strategies such as adjusting lighting or angles, changing between Turkish and English as target languages, and even turning to LLMs like ChatGPT for further clarification. This revealed that non-professional users of MT are not passive users of what current technology offers but creative problem-solvers who actively navigate the limitations of said technology.
“You have to know when to stop trusting it,” another student said. “If it’s something serious like police matters, I always ask a person afterwards.”
This pilot study suggests that trust in machine translation is highly context-sensitive, shaped by users’ lived experience, familiarity with the topic, and the perceived consequences of misunderstanding. For newly arrived students, these technologies often function as the first line of support in navigating unfamiliar systems. Yet they still require caution, critical judgment, and alternative strategies.
This emerging form of critical digital literacy, knowing how and when to rely on MT, and when not to, is becoming an essential survival skill in multilingual societies. For these students, machine translation was more than a helpful tool. It was a gateway to independence, participation, and daily understanding. Their personal reflections underscored how integral such tools have become to navigating life abroad, even while their shortcomings remain clear.
Sıla İlkılıç
The author is a graduate of Linguistic Data Sciences and is currently working as a project researcher for DECA.