Artificial intelligence tools have made significant advances in language translation, facilitating communication across diverse fields such as law, literature, and healthcare. However, experts caution that AI still faces considerable challenges when it comes to translating poetry, a domain that demands intricate cultural understanding and emotional nuance.

Poetry translation requires more than bilingual fluency; it involves a deep engagement with the source and target cultures, as well as the ability to interpret and convey metaphor, complex syntax, and mood. According to literary scholars, AI-powered translators often produce versions of poems that are overly standardized and lack the creative depth present in human translations.

To illustrate these limitations, Amherst College researcher Krupa Shandilya collaborated with poet and translator Adeeba Shahid Talukder to examine the challenges AI encounters when translating "Mulaqat" ("Meeting"), a 1953 poem by renowned Urdu poet Faiz Ahmed Faiz. Born in 1911 in pre-Partition India, Faiz was a prominent figure whose work melded classical poetic themes with pressing political issues of his time, including social justice and revolutionary ideals. His poetry’s ornate metaphors and emotional density make it notoriously difficult to translate accurately.

Faiz composed "Meeting" during a period of imprisonment from 1951 to 1955, when he was separated from his family and grappled with the impermanence of life and relationships. His experiences imbued the poem with layers of longing and melancholy that are challenging for AI to replicate. The poem centers on "vaṣl," the reunion of lovers, a traditional motif in Urdu poetry, but Faiz’s portrayal carries an acute awareness of loss and mortality.

One key example highlighted in the analysis involves the poem’s opening line, which the human translators rendered as: “This night is the dark, / lush tree of a grief greater / than you and I.” By contrast, an AI model translated it more literally as “This night is the tree of pain,” a version that, while not incorrect, misses the metaphor’s expansive emotional reach and complexity.

Further difficulties arise in AI's handling of the poem's layered syntax. The chatbot’s translation of a line describing “a thousand moons” weeping away their own light demonstrated an erroneous interpretation of reflexive pronouns in Urdu, leading to a confusing English rendering. This misstep underscores how AI struggles to parse nuanced grammatical structures that carry significant poetic weight.

Most notably, AI-generated translations fall short in conveying the poem’s emotional depth. Attempts to render metaphors like “dew of silence” resulted in awkward and incoherent phrasing, such as “From its dew in quiet,” diluting the original’s profound sense of sorrow and beauty. This shortcoming is partly attributed to AI’s lack of human experience, which limits its ability to authentically engage with the emotional texture of poetry.

Urdu literary critic Shamsur Rahman Faruqi has described Urdu poetic metaphor as a layered process where a metaphor becomes a factual base for further metaphoric elaboration. In the case of “Meeting,” the night as a metaphorical tree forms the foundation for imagery of dew as “drops of silence,” symbolizing grief turned into beauty—an intricate layering that AI translation currently cannot replicate.

While AI continues to transform practical communication and certain forms of literary translation, the art of capturing poetry’s full intricacy and emotional resonance remains, for now, a distinctly human endeavor.