English pragmatics tools — local collection
The open software for English pragmatics, collected 15.10.2026. Everything lives in pragmatics-tools (13 GB): src/ (62 git clones, depth 1), hf/ (9 HuggingFace models), manifest.tsv (kind, repo, detected licence, size), check.tsv, search.tsv (GitHub search hits by stars).
Licences. Per the build-vs-test rule, GPL, NC and repositories without a licence file (all rights reserved) are for tests and measurements only. MIT, Apache and BSD are candidates for building, after a check of the data they were trained on. What the tools mostly give us is the task definitions and the datasets they point to ; Mova's own pragmatics stays rule-based and transparent.
RST discourse parsing
repo licence use size
DMRST_Parser none (no licence file) tests only (no licence = all rights reserved) 6.2M isanlp_rst MIT License candidate (permissive) 3.3M DPLP MIT License candidate (permissive) 9.1M StageDP MIT License candidate (permissive) 33M rstfinder MIT License candidate (permissive) 12M feng-hirst-rst-parser see LICENSE.txt check the licence 236M rstWeb MIT License candidate (permissive) 6.3M DisCoDisCo none (no licence file) tests only (no licence = all rights reserved) 1.1M GumDrop Apache License candidate (permissive) 123M sharedtask2023 none (no licence file) tests only (no licence = all rights reserved) 571M sharedtask2021 none (no licence file) tests only (no licence = all rights reserved) 419M discourse-graphs see LICENSE check the licence 20M rst-converter Apache License candidate (permissive) 416K
PDTB / shallow discourse parsing, discourse relations
Discourse markers
repo licence use size
DiscSense none (no licence file) tests only (no licence = all rights reserved) 576K TSLM-DISCOURSE-MARKERS Apache License candidate (permissive) 1.3G
Coherence
repo licence use size
coherence-rst MIT License candidate (permissive) 388K GCDC-corpus none (no licence file) tests only (no licence = all rights reserved) 200K centering-neural-hds none (no licence file) tests only (no licence = all rights reserved) 2.5M DiscoScore none (no licence file) tests only (no licence = all rights reserved) 7.1M
Dialogue acts
Speech acts
repo licence use size
SGNN Apache License candidate (permissive) 24M
Politeness
repo licence use size
ConvoKit see LICENSE.md check the licence 27M politeness Apache License candidate (permissive) 5.0M politeness-classifier none (no licence file) tests only (no licence = all rights reserved) 332M
Irony and sarcasm
repo licence use size
tweeteval none (no licence file) tests only (no licence = all rights reserved) 27M sarcasm-detection MIT License candidate (permissive) 165M SarcasmDetection GNU LESSER GENERAL tests; linking only 12M
repo licence use size
metaphor-tsvetkov Open Database License check the licence 308M metaphor-in-context none (no licence file) tests only (no licence = all rights reserved) 432K MelBERT none (no licence file) tests only (no licence = all rights reserved) 2.2M MetaphorFrame none (no licence file) tests only (no licence = all rights reserved) 39M
Implicature
repo licence use size
ImpPres Attribution-NonCommercial tests only (NC) 12M pragmatic-implicature-llm none (no licence file) tests only (no licence = all rights reserved) 229M rsa-cgpotts GNU GENERAL PUBLIC LICENSE tests only (GPL) 12M pragmeval none (no licence file) tests only (no licence = all rights reserved) 21M ludwig MIT License candidate (permissive) 292K
Presupposition
repo licence use size
nope none (no licence file) tests only (no licence = all rights reserved) 468M CREPE BSD candidate (permissive) 412K
Rational Speech Acts
repo licence use size
webppl-rsa none (no licence file) tests only (no licence = all rights reserved) 276K
Certainty
Hedging
Stance
Pretrained models (HuggingFace, hf/)
Smoke test 15.10 on the GPU, two sentences each (transformers pipeline in venv ):
model task licence smoke test
cardiffnlp/twitter-roberta-base-ironyirony (TweetEval) not stated on the card "Great, another Monday. I just love being stuck in traffic." → irony 0.99; but "The train leaves at nine." → irony 0.72: over-eager outside tweets helinivan/english-sarcasm-detectorsarcasm (news headlines) not stated missed "Oh wonderful, the printer broke again." (not sarcastic, 0.87) Intel/polite-guardpoliteness, 4 classes Apache-2.0 "Could you please pass me the salt?" → polite 0.91; "Give me the salt now." → impolite 0.99 ✓ pedropei/sentence-level-certainty, aspect-level-certaintycertainty (Pei & Jurgens 2021) not stated a regression head: the classification pipeline output is meaningless; needs its own runner (src/certainty-certainty-estimator) sileod/roberta-base-discourse-marker-predictiondiscourse marker between two sentences (Discovery) Apache-2.0 "It was raining." / "we stayed home." → "mostly," 0.28 (expected "so"): weak diwank/silicone-deberta-pairdialogue acts (SILICONE) MIT not tested yet sileod/deberta-v3-base-tasksource-nli, cross-encoder/nli-deberta-v3-baseNLI, for implicature/presupposition probes (ImpPres, NOPE) Apache-2.0 not tested yet
Not found
probmods/problang (the RSA book code), hedgehog-nlp/hedgehog, vinodkpg/speech-act-classifier, tasnimmohiuddin/coherence-models: no such repositories (checked with git ls-remote).
Next
Nothing is wired into Mova. Candidates to measure against our pragmatics (prag), on the school and the pragmatics bench: polite-guard (requests, imperatives), discourse relations (discopy, DMRST — PDTB/RST over our trees), dialogue acts (SILICONE) for the speech-act frame, ImpPres/NOPE as implicature and presupposition tests.
← previous Expressions with their real meanings — the pragmatics lensnext → Mova RAG↑ back to docs This page in Markdown → https://md.mova.elamur.ai/free/en/docs/docs/pragmatics-tools/