Existing multilingual Bible corpora prioritize linguistic breadth - covering hundreds of languages with one or two versions each - but capture little of the translation depth available for major European languages. We introduce targum, a corpus of 657 New Testament translation instances (342 unique works) in English, French, Italian, Polish, and Spanish, aggregated from 13 sources and manually annotated with canonical work identifiers, revision years, and copyright status.

Alongside the texts, we provide chapter-level dense vector embeddings computed with Qwen3-Embedding-8B, which enable pairwise translation comparison, clustering, and benchmarking of new versions against 342 historical precedents. The corpus supports diachronic studies spanning 500 years of translation history, cross-lingual comparison of confessional traditions, and gradient stylistic analysis - research directions that breadth-focused resources cannot support. The public-domain subset and all embeddings will be made freely available; copyrighted texts will be available upon request.

I propose a case study where statistical estimation of the effect of political pressure on the literary translations seems possible. Translations of children’s books into Russian in the USSR during the period of the Cold War will serve as diagnostic data. The political alliances were well defined at the period, the iron curtain being a popular metaphor for the heavy state control of all cross-border communications, including those in the literary field. Quantitative analysis is enabled by the availability of a comprehensive bibliography. The key idea of the suggested measurement is that engagement with a foreign literary field is reflected in the publication of works by contemporary writers. Higher political pressure would mean less publications of contemporary works. Historical research on the Soviet translation highlights various reasons for a Western writer to be prohibited from publication in the USSR. Overall, authors who were still active throughout the period carried higher risks of being rejected for any reason compared to the authors of the past. Hence vitality coefficient—a proportion of contemporary works in all published translations—is indicative of the strength of the political censorship. Results show that vitality coefficient responded to the political alliances and hostility dynamics during the period in the predicted way: higher levels of confrontation led to lower share of the contemporary works in the translation flow. This fact helps realize a side effect of the political censorship barriers that is not much talked about: by impeding publication of current authors, censorship contributes to the consolidation of the position of the already well-established and familiar authors in the book market. If we assume that reprints contribute to the canonical status of a writer, censorship is then an indirect but real factor for canon consolidation.

Recent translation alignment algorithms based on multilingual sentence embeddings have achieved near-ideal accuracy on literary texts, enabling reliable detection of omissions and insertions. Yet from a translation studies perspective, omissions and insertions are only the most visible layer of divergence: beneath them lie the rewritings, compressions, register shifts, and pragmatic adaptations that transform a text while nominally preserving its content. This paper asks whether the same embeddings that power alignment can be repurposed as descriptive instruments for translation strategy.

We present a computational framework for characterizing literary translation beyond the binary of presence and absence, applied to a parallel corpus of ten translations of Alessandro Manzoni's "I Promessi Sposi" (1840) spanning eight languages, six language families, and 177 years (1845–2022), aligned at sentence level via TEI encoding. Two complementary metrics — LaBSE cosine similarity and expansion ratio — reveal that semantic fidelity and textual completeness are independent axes of translation strategy, producing a typology of ten distinct translator profiles ranging from faithful adaptor to free literary adaptor.

Cross-translation analysis reveals two patterns consistent across all ten translations regardless of language family or period, suggesting they reflect properties of Manzoni's source text rather than individual translator choice: the plague chapters (31–32) are universally the most divergent, and dialogue fidelity is consistently higher than narrator voice fidelity. Drift block analysis — combining low cosine similarity with compression ratio to detect consecutive runs of heavily compressed pairs — recovers systematic editorial programs invisible to aggregate metrics: Schröder's German translation (1879) conceals ten passage-level deletions targeting authorial digressions, interior monologues, and lyrical passages, while the anonymous Spanish translation (1858) concentrates its cuts in Gertrude's psychology and the plague historiography. Both translators independently reached the same passages by different editorial routes.