Poetry translation is the case the broader argument about translation cannot avoid. Prose can be paraphrased and survive. Poetry, if it survives at all, survives by being made again.
Three poetry translations published in late 2025 and early 2026 show three different working methods. None is the right way. Each is honest about what it gave up.
The first is Sasha Dugdale's revised version of Anna Akhmatova's Requiem, published by Bloodaxe Books in Hexham in November 2025. Dugdale is a poet first and a translator second, and her Akhmatova reads like a Sasha Dugdale poem. She would not contest this.
Her preface is candid. She kept Akhmatova's syllable counts where she could, abandoned them where she could not, and let the English line break find its own length. The result is an Akhmatova in English, not an English Akhmatova.
An older Akhmatova in English, the Stanley Kunitz and Max Hayward version of 1973, kept more of the Russian shape and less of the poem. The Judith Hemschemeyer complete poems, published by Zephyr in 1989 and revised in 1997, did the opposite of both, producing a Russian-flavoured English that has aged into something academic.
Dugdale's Requiem takes liberties. In the famous passage where the woman in the prison line asks Akhmatova, "Can you describe this?" and Akhmatova answers, "Yes, I can," Dugdale renders the answer as "Yes. I can do that." The added words are not in the Russian. They are, Dugdale argues, in the silence after the Russian, and English needs them.
The second case is Khaled Mattawa's English version of selected late poems by the Syrian poet Adonis, published by Yale University Press in February 2026. Mattawa, a Libyan-American poet and translator who has worked with Adonis for two decades, took the opposite approach.
Mattawa preserves the Arabic rhetorical structures. The repetitions, the apostrophic invocations, the long unspooling sentences that turn back on themselves. English does not naturally do these things. Mattawa makes English do them anyway.
His preface argues that the strangeness is the point. To smooth Adonis into idiomatic English is to translate the lines but not the poet. The reader of Mattawa's Adonis is meant to feel the Arabic pressing against the English.
The two cases are extreme opposites. Dugdale assumes the English-language reader wants a poem that works as a poem in English. Mattawa assumes the reader wants a poem that registers as a translation. Both are right about their particular reader. Neither is right in general.
The third case is the most interesting. In March 2026, Tilted Axis Press in London published a small volume of poems by the young Vietnamese poet Hà Mai Anh, translated by Quyên Nguyen-Le and Anna Moschovakis.
Hà Mai Anh is thirty-two and writes in a colloquial Hanoi Vietnamese that absorbs English internet phrases, classical Sino-Vietnamese, and a kind of street-market grammar that has no equivalent in any of the standard English registers.
The two translators worked together for eighteen months. Nguyen-Le produced literal versions that preserved the registers as code-switches. Moschovakis, who is a poet, then turned those code-switches into English where the registers shift just as abruptly.
The result is a book that reads, in English, as if it were written by someone who is impatient with English. Which is, Hà Mai Anh has said in interviews, what she wanted.
The book contains a glossary at the back. It is two pages long. It explains, for example, why the word chợ in one poem is translated as market and in another as scene. The two senses both exist in the Vietnamese. English will not carry both at once.
Anna Moschovakis, in a translator's note, says she came to think of the glossary as part of the poem. Hà Mai Anh agreed and asked for the glossary to be printed in the same font as the poems.
Tilted Axis did it.
What these three cases show, taken together, is that there is no single working theory of poetry translation that survives contact with the actual poem. Each book finds its method. Each method is partial.
The translator's job is to be honest about what was traded, and the reader's job is to read the translation as the new thing it is.
The Russian Akhmatova still exists. The Arabic Adonis still exists. The Vietnamese Hà Mai Anh still exists. The English versions are three more things in the world, and they are, on their own terms, worth having.
None of the three is the original. None of them claims to be. That refusal is, in 2026, the closest thing the field has to a working ethics.