Skip to content

data(WO67): tha-phonemic-v3 — wiki tier, dual-engine validated (130,906 train) - #347

Merged
ronaldtse merged 1 commit into
mainfrom
data/tha-phonemic-v3
Oct 22, 2026
Merged

data(WO67): tha-phonemic-v3 — wiki tier, dual-engine validated (130,906 train)#347
ronaldtse merged 1 commit into
mainfrom
data/tha-phonemic-v3

Conversation

@ronaldtse

@ronaldtse ronaldtse commented Oct 22, 2026

Copy link
Copy Markdown
Contributor

Closes the rule-composite self-expansion leg:

  • Tier: 61,561 types where royin and tltk (independent engines) agree, from 120k Thai Wikipedia articles (CC-BY-SA), frequency-filtered.
  • Validation: on 40,624 words shared with v2, the engines agree with v2's readings 100.0% of the time — engine independence validates the corpus pipeline end to end.
  • Contamination-safe: only the 20,937 novel types join train; v2's dev/test splits are byte-identical → v2-arm vs v3-arm is a controlled experiment.
  • Provenance: two deterministic engines + attributed text; no LLM teacher.

Ronald Tse
data(WO67): tha-phonemic-v3 — 130,906 train pairs, wiki tier dual-eng…
b62b663 
…ine validated

Wiki tier: 120k thwiki articles, frequency-filtered vocab, readings
from two independent deterministic engines; only dual-agreement rows
ship (61,561 of 61,953 types; 100.0% agreement on the 40,624 shared
words). Only novel types (20,937) join train — v2's dev/test are
byte-identical, so v2-arm vs v3-arm is a controlled comparison. No
LLM teacher anywhere.
@ronaldtse
ronaldtse merged commit 605cf03 into main Oct 22, 2026
@ronaldtse
ronaldtse deleted the data/tha-phonemic-v3 branch October 22, 2026 14:32
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@ronaldtse