📄 What Is Heard in the Mountains: Paul Celan’s Gespräch im Gebirg in the Light of its Hebrew Translation

Summary

This is a literary-studies essay, not an NLP paper. Ben‑Horin reads Paul Celan's prose piece Gespräch im Gebirg (Conversation in the Mountains) through its Hebrew translation by Shimon Sandbank. The story follows two Jewish cousins, Klein and Gross, who walk toward the mountains and never reach the summit. It is widely read as Celan's answer to Adorno's claim that poetry after Auschwitz is barbaric.

The author builds on Felman and Agamben, who see sound as central to testimony: Celan refuses a smooth, melodious language and uses a precise, broken one. She links the text to a planned meeting with Adorno in Sils‑Maria that never happened, to the Meridian speech, to Büchner's Lenz, and to Celan's remarks on "Judendeutsch", a deformed Jewish German.

The German original carries Yiddish syntax that Hebrew cannot reproduce. Sandbank compensated by adding Yiddish words that are absent from the source. The paper argues that such translator choices push readers to rethink how ethics and poetics relate in Celan's work.

Key points

• Venue: Yod 23 (2021), INALCO, pp. 185–202. The author is at Bar‑Ilan University. Digits in the extracted text are corrupted, so exact dates and ages cannot be reported reliably.
• Narrative structure: a third‑person opening shifts into a dialogue between "Du" (you) and "Ich" (I), and the text ends as a first‑person monologue. Almost no action takes place.
• Origin: Peter Szondi arranged a meeting between Celan and Adorno in Sils‑Maria (Engadin) that never took place. Celan turned this missed conversation into the text. A year later his Büchner Prize speech Meridian introduces ideas such as "Atemwende" (breathturn).
• Celan treated the jabbering "Judendeutsch" of the "crooked‑nose" figure as a path to the poem. He tied this to Günter Blöcker's review of Sprachgitter, which he read as antisemitic.
• Translation choice: in his epilogue, Sandbank describes inserting Yiddish lexical items into the Hebrew to stand in for the Yiddish syntax of the German.

Interesting ideas

• I1. Compensation in translation: when a source feature (here, Yiddish‑flavoured syntax) has no equivalent in the target language, mark the same register at another level, such as vocabulary, so the "otherness" of the source survives. — For Mova (back): when generating or translating through the meta-world, treat register and dialect as explicit graph attributes, and let the generator realise them by whatever means the target allows (lexicon instead of syntax).
• I2. Deviant language as signal, not noise: the paper treats broken, "disordered" speech as the carrier of meaning, not a defect to smooth away. — For Mova (tokens): keep nonstandard morphology and word order flagged as a register feature instead of normalising it, so later stages can still reason about who speaks and how.
• I3. Tracking perspective shifts: the text slides from third‑person narration to a you/I exchange and then to monologue, and reading it correctly depends on following who is speaking at each point. — For Mova (frames): add speaker/addressee tracking across narration-to-dialogue transitions, so that "I" and "you" are resolved to Klein or Gross rather than left as bare pronouns.
• I4. Translation as a diagnostic lens: the author uses the places where a translation diverges from its source to expose what the original was doing. — For Mova (learn): compare graphs extracted from a text and from its translation, and use mismatches as probes for which meanings the pipeline captures and which it loses.
• I5. Intertext as part of the world: much of the story's sense comes from echoes of Büchner's Lenz (walking through mountains, wishing to walk on one's head). — For Mova (world): store literary allusion links in the knowledge graph, so a reader module can explain an echo by pointing to its explicit source.