Millions of people begin their day by posting a friendly greeting on social media, never expecting that it could lead to their arrest. However, this is exactly what happened to a Palestinian construction worker in 2017. His Facebook selfie with the caption “??????” (“good morning”) was auto-translated as “attack them” by AI. This incident highlights the limitations of language technologies and the need for human language skills.
Advancements in automated translation have led some to believe that humans, especially English speakers, no longer need to learn other languages. They argue that apps like Google Translate can do the job for us. As a result, some universities are even considering dismantling their language programs.
However, machine language learning and human language learning are fundamentally different. Machines learn languages through algorithms trained on large amounts of text data. Bilingual training data, usually based on standard versions of languages, is used to train these algorithms. However, this approach fails to account for the diversity and variation present in human languages.
For example, the word “deadly” means “causing death” in most varieties of English. However, in Australian Aboriginal English, it means “excellent.” This diversity poses a challenge for machines, as they struggle to accurately translate such variations.
Furthermore, machines store languages differently than humans. Each language has its own unique characteristics and ways of encoding grammar. Translating a simple English statement like “I am a student” into German requires the inclusion of grammatical gender markings, resulting in different translations for male and female students.
Moreover, machine learning resources are heavily skewed towards English. Over 90% of the training data behind large language models is in English, while the majority of the world’s languages are underrepresented. This dominance of English further perpetuates inequalities in machine translation.
Machine translation is useful for getting the general idea of websites or asking for directions in a tourist destination. However, it falls short in high-stakes contexts like hospitals, where accurate communication is crucial. Translation apps can lead to errors and misunderstandings, putting patients at risk.
Therefore, it is essential to cultivate human multilingual talent. Only humans can assess the risks and determine when machine translation is appropriate. To make informed decisions, individuals need to understand both languages and machine learning.
Relying solely on machines for language tasks is costly and devalues the importance of language learning. By outsourcing language work to machines, we miss out on developing advanced language proficiency and the ability to communicate effectively in complex situations.
Languages are diverse, complex, and deeply social, while algorithms are not. Embracing the idea that machines can replace human language skills dehumanizes the essence of language as a means of communication, meaning-making, relationship-building, and community-building.
