alc-english/pimsleur-study/ANALYSIS.md

Pimsleur Spanish — Tape Format, Activities & Methodology

ASR study of the Pimsleur Spanish course (levels I–V) to establish the tape format, the activities present, and the underlying methodology.

Corpus

Source: /home/j/projects/Pimsleur Spanish/ (~4.5 GB):

Level	Units	Notes
Spanish I	30	+ separate Reading Lessons track (booklet-based)
Spanish II	30
Spanish III	30
Spanish IV	~26
Spanish V	~30 (51 files)	extra material

Each unit ≈ 30 min (~30 MB MP3). ~85 hours total.

Transcribed with Soniox stt-async-v4 (es+en hints), with the long learner pauses annotated (they encode the "your turn" windows). Sample read for this write-up: I-01, I-02, I-15, I-30; II-01; III-01; IV-01; V-01; Reading-01. Transcripts live in pimsleur-study/transcripts/.

Unit format (the ~30-minute "tape")

Consistent across all levels:

1. Header + target dialogue, cold — "This is Unit N of Pimsleur's Spanish L. Listen to this conversation." then the native-speed Spanish dialogue. [I-01 00:02]
2. Dialogue replay — "Listen again / Escuche otra vez." [II-01 00:41]
3. Core drill body (~25 min) — a continuous stream of recall prompts that (a) rebuild the dialogue piece by piece and (b) interleave previously-learned material at expanding intervals.
4. Final replay + role-play — "Now imagine you are the American…"; the learner performs one side of the conversation from cues. [I-01 26:36]
5. Close — mastery gate ("If you've mastered about 80%… move on"), "This is the end of Unit N." [I-01 29:36]. The last unit of a level hands off to the Reading track. [I-30 29:18]

Activities present

• A. Listen-to-dialogue — comprehension; cold open + replays.
• B. New-item intro: Listen & Repeat with backward buildup — a phrase is taught from the END backward to preserve intonation: "Don" → "Per" → "Perdón." [I-01 00:55–01:12]
• C. Pronunciation coaching — explicit micro-notes: the Spanish r, vowel quality, silent h, double-ll = /y/, stress on the next-to-last syllable. [I-01 02:22; Reading-01 03:13]
• D. Graduated-Interval Recall prompts (the core engine) — instruction → silent pause (learner produces aloud) → native confirmation. Prompt forms:
  • "How do you say X?" / "Say X."
  • "Ask: Do you…?" / "Answer: …"
  • English meaning → Spanish production ("I don't speak English." → pause → "No hablo inglés.") Items reappear at lengthening intervals (spaced retrieval).
• E. Recombination / transformation — assemble new utterances from known parts; reorder for questions ("reverse the first two words to ask 'Are you American?'"). [I-01 23:34]
• F. Number / set drills — counting, telling time, woven into the same prompt loop. [I-15 01:13+]
• G. Role-play / anticipation — "you are X; respond" multi-turn simulated conversation. [IV-01 01:36]
• Reading track (separate) — phonics/decoding from a booklet: read printed words aloud after the model with spelling→sound rules; word → phrase progression. Optional, "one reading after every other unit, or at the end." [Reading-01]

Methodology (Pimsleur principles, with evidence)

• Graduated Interval Recall — expanding-interval spaced retrieval is the spine of every unit.
• Principle of Anticipation — the learner is cued to produce before hearing the answer (active recall, not echoing), enforced by the timed pause + confirmation.
• Backward buildup for new phrases (keeps natural prosody).
• Audio-only core, minimal explicit grammar — patterns are induced from examples; grammar is named only in passing ("question intonation", "literally, in a trip of business"). Reading/writing is fully separated into the optional Reading track.
• Tight core of high-frequency vocabulary, recycled relentlessly.
• L1 → L2 instruction fade — the language of instruction shifts from English (L1) toward Spanish by L4 ("Trate de decir…", "Contéstele", "¿A qué se dedica?"); English is kept only to specify the meaning to produce. [IV-01, V-01]
• One 30-min unit/day, strictly sequential; ~80% mastery to advance.

Progression across levels

• L1 — single words → short phrases; English narrator; heavy scaffolding + method explanation. Topics: greetings, understand/speak, numbers, time. [I-01, I-15]
• L2 — fuller social dialogue; present perfect ("he estado", "he hablado"); narrator starts instructing in Spanish ("¿Se acuerda de…?", "Dígale…"). [II-01]
• L3 — longer multi-turn dialogues (dinner party); past/future, weather, travel; instructions largely Spanish. [III-01]
• L4 — near-native Spanish instruction; full-sentence production with long (7–8 s) pauses; professional/travel scenarios. [IV-01]
• L5 — extended scenarios (gifts, family, shopping); subjunctive ("algo que tenga que ver con México"); more material (~50 files). [V-01]

Data-quality notes

• Soniox handles the bilingual audio well; word starts are reliable (ends are not — irrelevant here).
• This model didn't return per-token language, so the (??) tags in the .txt are not split EN/ES (the text itself makes the language obvious). If we want clean EN/ES segmentation we can post-classify per word by script + a small lexicon.
• Pause markers approximate the learner-response windows and make the prompt → produce → confirm rhythm legible.

Implications for our content (ALC / module player)

• Pimsleur's prompt → timed pause → confirmation maps almost 1:1 onto our PRODUCE activity (prompt → respond window → model response) and the repeat/respond cues already built into the player/video.
• Backward buildup and graduated-interval recall scheduling are concrete, portable mechanics we could add as activity types / a scheduler.
• The L1→L2 instruction fade is a design lever: progressively localize instructions into the target language as the learner advances.