Reading academic papers in a second language without losing momentum

Reading research papers is hard enough in your native language. Reading them in a second language adds a tax that can compound into hours per paper. Researchers, PhD students, and professionals consuming foreign-language scholarship all hit a version of the same problem: the productive workflow for understanding the paper conflicts with the productive workflow for building vocabulary that helps you read the next paper faster.

Most readers solve this conflict by sacrificing the vocabulary side. They rush through papers, miss subtleties, look up only the most blocking words, and never come back. Five years later they're still slow at reading papers, still rely heavily on translation tools, and still feel like an outsider in the academic discourse.

This article describes a workflow that builds reading speed AND domain vocabulary in parallel, optimised for the realities of academic reading: dense text, field-specific terminology, recurring constructions, time pressure.

The two-pass reading method

The single biggest improvement most second-language paper readers can make is separating comprehension from vocabulary capture.

Pass 1 — Comprehension only (40-60 min for a typical paper)

Your goal: understand the paper's claims, methods, and conclusions. Vocabulary is a means, not an end.

Read the abstract, introduction, and conclusion first.
Skim the methods section to understand what they did.
Read results focusing on figures.
Read discussion for interpretation.
For unfamiliar words: look them up briefly to maintain comprehension, but don't pause to capture them.

Use the dictionary aggressively. Tools like Mnemo's Chrome extension let you see definitions in a popup without leaving the page — minimum disruption.

When you finish Pass 1, you should be able to summarise the paper in 3 sentences. If you can't, you missed something — re-read targeted sections.

Pass 2 — Vocabulary capture (15-25 min)

Now you're not reading for comprehension; you're scanning the paper for vocabulary worth saving. This is faster than Pass 1 because you already understand the content.

For each unfamiliar word you noticed:

Decide: is this field vocabulary (likely to recur in similar papers) or incidental vocabulary (used only here)?
Capture only field vocabulary. Use the highlight-to-save flow.
For each captured word, save the enclosing sentence — context is everything for academic vocabulary.

For a typical 12-page paper, expect to capture 15-30 field-vocabulary items. Skip:

Words that appear once and are obviously incidental.
Proper nouns (names, places, software).
Words you can guess from context and unlikely to recur.

Quality over quantity. Better to have 200 captured words you actually use than 2,000 you've forgotten.

What counts as "field vocabulary"

Within any academic field, there's a layered vocabulary structure:

Common academic English (~3,000 words): substantiate, paradigm, hypothesise, concomitant, empirical, robust. Universal across fields. The Coxhead Academic Word List is a good starting set.
Field-specific terminology (~500-2,000 words per field): in machine learning — embedding, backpropagation, transformer, fine-tuning. In molecular biology — transcription, epigenetic, chromatin. In economics — elasticity, equilibrium, exogenous.
Sub-field jargon (~100-500 words): in NLP — attention head, causal mask, RoPE, KV cache. Specific to a sub-community.
Recurring phrasing patterns — not exactly vocabulary but worth capturing: "we hypothesise that...", "this is consistent with...", "in line with prior work...", "controlling for...".

For a researcher new to a field: focus initial capture on layers 2 and 4. These give the highest return — once you know the field-specific terms and the recurring patterns, you read papers in the field 2-3× faster.

Domain decks

Maintain separate flashcard decks per domain you read in. Don't mix machine learning vocabulary with neuroscience vocabulary in one deck — even if they share some words, the contexts differ.

Suggested deck structure for an ML researcher:

Common Academic — Coxhead-list-style words from any paper.
ML Foundational — terms from undergraduate-level ML textbooks, recurrent across the field.
NLP (or your sub-field) — terms specific to your sub-area.
Recent Papers — vocabulary from this month's papers, 6-month rolling window.

Review all four mixed via FSRS. The interleaving (mixing words from different decks within one review session) actually improves retention because your brain has to work harder to discriminate.

A typical research session

Concrete example, ~2 hours:

00:00-00:50  Pass 1 — Read paper for comprehension. 
             Browser-popup dictionary on for blockers.
             No saving.

00:50-01:00  Synthesis — Write 3-sentence summary in your own words. 
             Note the paper's contribution. Bookmark or cite as needed.

01:00-01:25  Pass 2 — Re-scan for vocabulary. 
             Capture field-specific terms via extension.
             ~20 words captured.

01:25-01:50  Curate — Open inbox in web app. 
             For each new card: add 1 collocation, write a context-relevant 
             sentence, tag with domain.

01:50-02:00  Review — Daily FSRS review across all domain decks.

After 50 papers (1-2 months for active researchers), you have 1,000-1,500 well-curated field-vocabulary items reviewed via FSRS. At that point, paper-reading speed improves dramatically — you stop re-looking-up the same recurring terms.

Common research-reading mistakes

Trying to capture everything in Pass 1

Reading + capturing simultaneously is cognitively expensive. Comprehension suffers, capture quality suffers, you finish exhausted and didn't actually learn the paper.

Not saving enclosing sentences

A word in isolation is hard to remember and hard to use. Concomitant in a list is meaningless; concomitant in "we observed a concomitant decrease in expression of related genes" is concrete.

Skipping curation

Captured-but-not-curated cards are noise. The 10 minutes of curation per paper is what makes the system work; without it, you're back to a wishlist that you never actually study.

Giving up on monolingual definitions

For C1+ readers, switching to monolingual definitions in your second language strengthens in-language thinking. Translation back to your native language adds a step that you don't want when reading at speed. Set your dictionary preference accordingly.

Ignoring recurring phrasing patterns

Academic English has highly stylised patterns. Capturing 30 "academic phrasing" cards (it has been argued that..., contrary to received wisdom..., we propose to...) is one of the highest-leverage things a second-language researcher can do. Most researchers never explicitly study these.

Tools and setup

For this workflow specifically, you want:

Browser-based dictionary popup that doesn't leave the page (Pass 1 lookup).
Highlight-to-save Chrome extension (Pass 2 capture).
Multi-deck flashcard tool with FSRS.
Sync across devices — you'll review on phone in commute time.
Markdown notes — for paper summaries and connecting concepts.

Mnemo provides all of this in one tool — Chrome extension for in-page lookup and capture, web app with FSRS-scheduled review and Markdown journal for summaries. Free, multi-language. Try free.

Anki + a community PDF reader add-on can achieve a similar workflow if you're already an Anki power user. Native PDF readers (Adobe Acrobat, Skim) generally have weak SRS integration.

Long-term outcome

Researchers who run this workflow consistently for 6-12 months report:

Paper reading speed improving 1.5-2× (from baseline of ~2 hrs/paper to ~1 hr/paper for typical field papers).
Confidence in writing papers in the second language — they have the academic phrasing internalised.
Reduced reliance on translation tools — comprehension shifts toward direct in-language reading.
The boundary between "reading for understanding" and "reading for vocabulary" disappears, because vocabulary is being absorbed naturally as a byproduct of disciplined reading.

This isn't unique to a specific tool — Anki users running similar workflows report the same outcomes. What matters is the two-pass discipline + deliberate field/incidental separation + curation step. The tool is downstream.

Summary

Reading academic papers in a second language is a long-term skill. The workflow:

Two-pass reading — comprehension first, capture second.
Capture only field vocabulary — be ruthless about incidental words.
Domain-separated decks — but reviewed mixed via FSRS.
Curate immediately after reading — collocations, context sentences, tags.
Capture phrasing patterns, not just words.
Daily FSRS review across all domain decks.

Six months of this beats five years of "I'll look it up next time".