Writing study: Simultaneous Speech-to-Speech Translation Without Aligned Data

ICML overviewCorpus atlasBy contribution typeVisual galleryRules & templateWriting patternsPaper studies

← ICML 2026 oral writing corpus

Tom Labiausse, Romain Fabre, Yannick Estève, Alexandre Défossez, Neil Zeghidour · Official ICML 2026 oral record

Observed writing choices, not acceptance rules. This study examines the inspected manuscript version. It does not establish why the paper was accepted or that its presentation choices should be copied unchanged.

Figure, table and section ledger

Editorial contribution type: Training, objectives & optimization (primary).

Layout extraction, not a required paper template. Counts are distinct caption labels; a multi-panel figure counts once. Section spans are estimated from heading positions, including floats and whitespace. Method and limitations.

Canonical PDF

20 physical pages

Before references / appendix

≈ 8.66 page-equivalents

Main figures / tables

8 / 3

Attached appendix figures / tables

6 / 3

Attached appendix span

≈ 7.91 page-equivalents

Before references / appendix: ≈ 43.3%References / intervening material: ≈ 17.15%Attached appendix: ≈ 39.55%Estimated layout envelopes, not text/artwork area. Unresolved appendix boundaries remain within the undivided post-body material.

Section-by-section lengths and source pages

The title/abstract/front matter precede the first numbered heading. A span ends at the next detected top-level heading, not at the last sentence; end-page whitespace is included. References and statements are listed separately. These are not measurements of text-only or artwork area.

Top-level heading	Starts on PDF page	Estimated span	Share of entire PDF
1. Introduction	1	≈ 0.74 pp.	3.7%
2. Related Work	2	≈ 0.67 pp.	3.35%
3. Method	2	≈ 2.6 pp.	13%
4. Experiments	5	≈ 4 pp.	20%
5. Conclusion	9	≈ 0.2 pp.	1%
Impact Statement	9	≈ 0.12 pp.	0.6%
References	9	≈ 3.42 pp.	17.1%
Appendix	13	≈ 7.91 pp.	39.55%

All detected figure and table locations

Main text

• Figure 1 · p. 3
• Figure 2 · p. 3
• Figure 3 · p. 5
• Table 1 · p. 6
• Table 2 · p. 7
• Table 3 · p. 7
• Figure 4 · p. 7
• Figure 5 · p. 8
• Figure 6 · p. 8
• Figure 7 · p. 8
• Figure 8 · p. 9

Attached appendix

• Table 4 · p. 13
• Table 5 · p. 14
• Table 6 · p. 14
• Figure 9 · p. 15
• Figure 10 · p. 16
• Figure 11 · p. 17
• Figure 12 · p. 18
• Figure 13 · p. 19
• Figure 14 · p. 20

Argument and organization

1. 1 Introduction; 2 Related Work

   Frames fine-grained alignment as a training bottleneck for simultaneous translation.

2. 3 Method

   Defines multistream modeling, coarse alignment and the reinforced translation policy.

3. 4 Experiments; 5 Conclusion

   Orders protocol, metrics, comparative results and ablations before concrete deployment limitations.

Problem, claim and contribution

The abstract narrows the title’s no-alignment claim to eliminating word-level alignment while retaining sentence-level paired supervision.

Abstract and Introduction, p.1 · Main paper

Argument and section order

Architecture, data alignment and policy optimization are distinct method components; the experiment section separately enumerates training stages and inference configuration.

Section 3, pp.2-5; Sections 4.2 and 4.5, pp.5-7 · Main paper

Figures, captions and visual explanation

A worked bilingual waveform example anchors intermediate BLEU rewards to a marked time boundary, translating the reward definition into an observable timing difference.

Figure 3, p.5 · Main paper

Tables, comparisons and evidence

Human audio-quality/similarity/naturalness scores are kept separate from objective translation/latency measures; the adaptation table includes baseline, fine-tuned and reinforced stages.

Tables 2-3 and evaluation prose, p.7 · Main paper

Appendix and reproducibility

The supplementary multilingual-versus-monolingual table explains unequal total training exposure and supplies both equal-update and equal-per-language exposure comparisons.

Appendix Table 4, p.13 · Main paper

Clarity and writing conventions

The limitations state that accent strength and the quality-latency trade-off cannot currently be adjusted at inference, identifying specific missing controls rather than generic future work.

Section 4.8, p.9 · Main paper

What to reuse—and what not to infer

Transferable lesson: Qualify a broad title in the opening and separate perceptual quality, task accuracy and latency with explicit metric and supervision contracts.

Boundary: No word-level alignment does not mean unpaired training; four source-to-English settings and one adaptation study do not establish unrestricted multilingual transfer.

Apply this lesson using the argument and evidence worksheet.

Inspection record and sources

Checked

2026-10-07

Reading status

Writing-focused manuscript scan + selected visuals inspected

Main text

Writing-oriented scan of the main argument through physical p.9, with focused physical-page text inspection on 1, 3, 5, 7, 9. Framing, section progression, mechanism/proof explanation, displays, comparison conditions and conclusions were inspected. This is not exhaustive technical reading, proof verification or reproduction.

Appendix / supplement

Appendix organization and main-text pointers were checked; focused text inspection of physical pages 13. Remaining appendix proofs and details were not exhaustively read. No separately linked supplement inspected.

Visually inspected PDF pages

5, 7

This is a writing-focused scan of the manuscript and the stated supplementary scope: framing, organization, visual explanation, evidence presentation and wording. It is not a line-by-line proof audit, reproduction of results or exhaustive scientific review.

• Published proceedings PDF, PMLR volume 306, retrieved 2026-10-07 · 20 pages