SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2603.00620 · 2026

QQ: A Language Metadata Toolkit for Multilingual NLP

Miryam De Lhoneux, Yiyi Chen, Wessel Poelman, L O M•nlp

arXiv · PDF · Open in the Atlas

Code that ran

17 of 19 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 2 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 15 failures of our instrument; 2 no recorded run), measured 2026-11-22; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2603.00620.json · badge: /api/ran/2603.00620.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
WPoelman/qwanqwa canonical (classifier label)
classifier basis: census
15 ran · 1 no recorded run
pycountry/pycountry canonical (classifier label)
classifier basis: census
1 ran · 0 no recorded run
rspeer/langcodes community
Papers with Code archive
1 ran · 1 no recorded run
FunctionStatusWhere it lives
_get2 Ran
ran; our contract draft was wrong, not the code
rspeer/langcodes/langcodes/language_distance.py
code served (permissive licence) · get_code("87ef13f56a39c168")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Construct a simple nested dictionary structure for testing _get2
dictionary = {
    "level1_key": {
        "level2_key": "found_value"
    },
    "another_key": {
        "nested": "other_value"
    }
}

# Set up arguments for _get2
# The function performs: dictionary.get(key1, {}).get(key2, default)
# It tries to get key1 from dictionary (defaulting to empty dict if not found)
# Then tries to get key2 from that result (defaulting to the provided default value)

ARGS = {
    "dictionary": dictionary,
    "key1": "level1_key",
    "key2": "level2_key",
    "default": "default_value"
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
build_colexification_graph Ran
ran; our contract draft was wrong, not the code
WPoelman/qwanqwa/case-studies/linking-datasets/colexification_graph.py
code served (permissive licence) · get_code("6b629ba8d05d27b2")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Graph. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import networkx as nx
from itertools import combinations

# Build a minimal concepts dictionary
concepts = {
    'c001': {
        'gloss': 'a domesticated mammal',
        'synset_id': 'synset_001',
        'n_babel_langs': 42
    },
    'c002': {
        'gloss': 'a large domesticated mammal',
        'synset_id': 'synset_002',
        'n_babel_langs': 38
    },
    'c003': {
        'gloss': 'a young dog',
        'synset_id': 'synset_003',
        'n_babel_langs': 35
    },
    'c004': {
        'gloss': 'a feline animal',
        'synset_id': 'synset_004',
        'n_babel_langs': 40
    },
}

# Build a minimal index dictionary
# Key: (language, lemma_lowercase)
# Value: list of (concept_id, original_lemma) tuples
index = {
    ('English', 'dog'): [('c001', 'dog'), ('c002', 'dog')],
    ('English', 'puppy'): [('c003', 'puppy')],
    ('French', 'chien'): [('c001', 'chien'), ('c003', 'chien'), ('c004', 'chien')],
    ('Spanish', 'gato'): [('c004', 'gato'), ('c002', 'gato')],
    ('German', 'katze'): [('c004', 'katze')],
}

ARGS = {
    'concepts': concepts,
    'index': index,
    'max_polysemy': 5
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
networkx — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Graph — a Graph output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Graph output the recorded ndim 1 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
collect_codes Ran
ran; our contract draft was wrong, not the code
WPoelman/qwanqwa/case-studies/linking-datasets/link.py
code served (permissive licence) · get_code("847c83bc6da9a368")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned set. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import tempfile
import zipfile
import csv
from pathlib import Path

# Create a temporary zip file with a CSV inside
temp_dir = tempfile.TemporaryDirectory()
temp_path = Path(temp_dir.name)
zip_path = temp_path / "test.zip"

# Create a zip file with a CSV file inside
with zipfile.ZipFile(zip_path, 'w') as zf:
    csv_content = "code\tname\tdescription\n"
    csv_content += "ABC123\tItem A\tFirst item\n"
    csv_content += "DEF456\tItem B\tSecond item\n"
    csv_content += "ABC123\tItem C\tDuplicate code\n"
    csv_content += "GHI789\tItem D\tThird item\n"
    zf.writestr("data.csv", csv_content)

ARGS = {
    "zip_path": zip_path,
    "inner_path": "data.csv",
    "col_names": ["code", "name"],
    "sep": "\t"
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type set — a set output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a set output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
collect_codes_etymon Ran
ran; our contract draft was wrong, not the code
WPoelman/qwanqwa/case-studies/linking-datasets/link.py
code served (permissive licence) · get_code("a11bffd935e64e99")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned set. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import io
import zipfile
import tempfile
from pathlib import Path

# Create a temporary zip file with valid Etymon TSV data
temp_dir = tempfile.TemporaryDirectory()
temp_path = Path(temp_dir.name)
zip_file_path = temp_path / "etymon_data.zip"

# Create sample TSV content in the format: {lang}: {lemma}\t{relation}\t{lang}: {lemma}
tsv_content = """en: water\trelated\tfr: eau
de: wasser\tetymology\tnl: water
es: agua\tcognate\tit: acqua
pt: água\tsimilar\tro: apă"""

# Create the zip file with the TSV content
with zipfile.ZipFile(zip_file_path, 'w') as zf:
    zf.writestr("etymon_data.tsv", tsv_content)

ARGS = {
    "zip_path": zip_file_path,
    "inner_path": "etymon_data.tsv"
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type set — a set output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a set output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
collect_codes_phonotacticon Ran
ran; our contract draft was wrong, not the code
WPoelman/qwanqwa/case-studies/linking-datasets/link.py
code served (permissive licence) · get_code("3bb7ae9d1e8b44a4")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import csv
import io
import zipfile
from pathlib import Path
import tempfile

# Create a temporary zip file with CSV data containing Glottocodes and ISO codes
temp_dir = tempfile.mkdtemp()
zip_path = Path(temp_dir) / "test.zip"

# Create CSV content with sample data
csv_content = """Glottocode,ISO,Language
abcd1234,eng,English
efgh5678,fra,French
ijkl9012,deu,German
"""

# Create the zip file with the CSV inside
with zipfile.ZipFile(zip_path, 'w') as zf:
    zf.writestr("phonotacticon.csv", csv_content)

# Create a minimal resolver (not used in the function, but required as parameter)
class MinimalResolver:
    pass

resolver = MinimalResolver()

ARGS = {
    "zip_path": zip_path,
    "inner_path": "phonotacticon.csv",
    "resolver": resolver
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 1, dtype family float — a tuple output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
format_languoid_structure Ran
ran; our contract draft was wrong, not the code
WPoelman/qwanqwa/src/qq/release.py
code served (permissive licence) · get_code("92f5f4720885c343")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

class MockLanguoid:
    def __init__(self):
        self.name = "English"
        self.iso_639_3 = "eng"
        self.bcp_47 = "en"
        self.iso_639_2b = "eng"
        self.glottocode = "stan1293"
        self.wikidata_id = "Q1860"
        self.endonym = "English"
        self.speaker_count = 1500000000
        self.level = MockLevel("language")
        self.scope = "individual"
        self.status = "living"
        self.is_macrolanguage = False
        self.endangerment_status = MockEndangermentStatus("not_endangered")
        self.script_codes = ["Latn", "Arab"]
        self.parent = MockLanguoid_Parent()
        self.family_tree = [MockLanguoid_Ancestor(), MockLanguoid_Ancestor2()]
        self.children = [
            MockChild("English Creole 1"),
            MockChild("English Creole 2"),
            MockChild("English Creole 3"),
            MockChild("English Creole 4"),
            MockChild("English Creole 5"),
            MockChild("English Creole 6"),
        ]
        self.scripts = [
            MockScript("Latin", "Latn"),
            MockScript("Arabic", "Arab"),
        ]
        self.canonical_scripts = self.scripts
        self.regions = [
            MockRegion("United Kingdom", "GB"),
            MockRegion("United States", "US"),
            MockRegion("Canada", "CA"),
            MockRegion("Australia", "AU"),
            MockRegion("Ireland", "IE"),
            MockRegion("New Zealand", "NZ"),
        ]
    
    def nllb_codes(self, use_bcp_47=False):
        if use_bcp_47:
            return ["en_Latn", "en_Arab"]
        return ["eng"]
    
    def descendants(self):
        return [MockChild(f"Descendant {i}") for i in range(42)]


class MockLevel:
    def __init__(self, value):
        self.value = value


class MockEndangermentStatus:
    def __init__(self, value):
        self.value = value


class MockLanguoid_Parent:
    def __init__(self):
        self.name = "Indo-European"
        self.iso_639_3 = None
        self.glottocode = "indo1319"


class MockLanguoid_Ancestor:
    def __init__(self):
        self.name = "Proto-Indo-European"
        self.level = MockLevel("proto-language")


class MockLanguoid_Ancestor2:
    def __init__(self):
        self.name = "Germanic"
        self.level = MockLevel("language_family")


class MockChild:
    def __init__(self, name):
        self.name = name
        self.iso_639_3 = "xxx"
        self.glottocode = "xxxx1234"


class MockScript:
    def __init__(self, name, iso_15924):
        self.name = name
        self.iso_15924 = iso_15924


class MockRegion:
    def __init__(self, name, country_code):
        self.name = name
        self.country_code = country_code


ARGS = {
    "languoid": MockLanguoid()
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
lazy_load Ran
ran; our contract draft was wrong, not the code
pycountry/pycountry/src/pycountry/db.py
pointer only (licence: LGPL-2.1) · get_code("f850e5c174c44a84")
How it ran

The body of this sample is withheld: LGPL-2.1, as GitHub detects it on this paper's repository, pycountry/pycountry, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

load_babelnet Ran
ran; our contract draft was wrong, not the code
WPoelman/qwanqwa/case-studies/linking-datasets/colexification_graph.py
code served (permissive licence) · get_code("0d082022d3ceb3b3")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import csv
import io
import zipfile
from collections import defaultdict
from pathlib import Path
import tempfile
import os

# Create a temporary zip file with a valid BabelNet TSV structure
temp_dir = tempfile.mkdtemp()
zip_path = Path(temp_dir) / "babelnet.zip"
tsv_filename = "babelnet.tsv"

# Create TSV content
tsv_content = """concept_id	language	lemma	source	concept_gloss	synset_id
bn_001	en	apple	WIKT_1	a fruit	wn30_i1234
bn_001	en	apple fruit	WIKT_2	a fruit	wn30_i1234
bn_001	fr	pomme	OMWN_1	un fruit	wn30_i1234
bn_002	en	dog	WN_1	a canine animal	wn30_i5678
bn_002	en	canine	MCR_1	a canine animal	wn30_i5678
bn_002	de	Hund	OEWN_1	ein Hund	wn30_i5678
bn_003	en		WIKT_3	empty lemma test	wn30_i9999
bn_004	en	under_score	SLOWNET_1	lemma with underscore	wn30_i2222
bn_005	en	table	INVALID_SOURCE_1	furniture	wn30_i3333
bn_005	en	table	WN_2	furniture	wn30_i3333
bn_006	en	run	WIKT_1	to move fast	wn30_i4444
bn_006	en	run	WIKT_2	to move fast	wn30_i4444
bn_006	en	run	WIKT_3	to move fast	wn30_i4444
bn_006	en	run	WIKT_4	to move fast	wn30_i4444
bn_006	en	run	WIKT_5	to move fast	wn30_i4444
bn_006	en	sprint	WIKT_6	to move fast	wn30_i4444
"""

# Create zip file with the TSV
with zipfile.ZipFile(zip_path, 'w') as zf:
    zf.writestr(tsv_filename, tsv_content)

ARGS = {
    "zip_path": zip_path,
    "inner_tsv": tsv_filename,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 1, dtype family float — a tuple output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
load_concepticon Ran
ran; our contract draft was wrong, not the code
WPoelman/qwanqwa/case-studies/linking-datasets/concept_graph.py
code served (permissive licence) · get_code("6621368410a34f5d")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import csv
import io
import zipfile
from pathlib import Path
from collections import defaultdict
import tempfile

# Create a temporary zip file with TSV data
temp_dir = tempfile.TemporaryDirectory()
zip_path = Path(temp_dir.name) / "concepticon.zip"

# Create sample TSV content
tsv_content = """concept_id\tlanguage\tlemma\tconcept_gloss\tsynset_id
c001\ten\tdog\tcanine animal\tss001
c001\tde\tHund\tcanine animal\tss001
c002\ten\tcat\tfeline animal\tss002
c002\tfr\tchat\tfeline animal\tss002
c002\ten\tkitty\tfeline animal\tss002
"""

# Write the zip file with TSV inside
with zipfile.ZipFile(zip_path, 'w') as zf:
    zf.writestr('concepticon.tsv', tsv_content)

ARGS = {
    'zip_path': zip_path,
    'inner_tsv': 'concepticon.tsv'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
load_ratings Ran
ran; our contract draft was wrong, not the code
WPoelman/qwanqwa/case-studies/linking-datasets/concept_graph.py
code served (permissive licence) · get_code("75d6a7f5a07a386a")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import csv
import io
import zipfile
from collections import defaultdict
from pathlib import Path
import tempfile
import os

# Create a temporary zip file with TSV data for testing
temp_dir = tempfile.mkdtemp()
zip_path = Path(temp_dir) / "test_data.zip"

# Create sample TSV data
tsv_content_1 = "CONCEPTICON_ID\trating_col1\trating_col2\tother_col\n"
tsv_content_1 += "c001\t3.5\t4.2\tsome_value\n"
tsv_content_1 += "c002\t2.1\t3.8\tanother_value\n"

tsv_content_2 = "CONCEPTICON_ID\trating_col1\n"
tsv_content_2 += "c001\t2.5\n"
tsv_content_2 += "c003\t1.9\n"

# Create zip with proper directory structure
with zipfile.ZipFile(zip_path, 'w') as zf:
    zf.writestr('datasets/dataset1/data.tsv', tsv_content_1)
    zf.writestr('datasets/dataset2/data.tsv', tsv_content_2)

# Create var_meta dict with metadata for rated columns
var_meta = {
    ('dataset1', 'rating_col1'): {'type': 'semantic_similarity', 'lang': 'en'},
    ('dataset1', 'rating_col2'): {'type': 'semantic_similarity', 'lang': 'en'},
    ('dataset2', 'rating_col1'): {'type': 'relatedness', 'lang': 'fr'},
}

ARGS = {
    'zip_path': zip_path,
    'var_meta': var_meta,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
load_variable_metadata Ran
ran; our contract draft was wrong, not the code
WPoelman/qwanqwa/case-studies/linking-datasets/concept_graph.py
code served (permissive licence) · get_code("dea05c7553d830d5")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import csv
import io
import zipfile
from pathlib import Path
from collections import defaultdict
import tempfile

# Create a temporary zip file with a valid TSV inside
def create_test_zip():
    temp_dir = tempfile.mkdtemp()
    zip_path = Path(temp_dir) / "test.zip"
    
    # Create TSV content
    tsv_content = """DATASET\tNAME\tTYPE\tLANGUAGE\tNORARE
dataset1\tCOL_VALENCE_MEAN\tvalence\tenglish\tratings
dataset1\tCOL_VALENCE_MIN\tvalence\tenglish\tratings
dataset1\tCOL_AROUSAL_MEAN\tarousal\tenglish\tratings
dataset2\tSPANISH_AOA_MEAN\tAoA\tspanish\tnorms
dataset2\tSPANISH_AOA_MAX\tAoA\tspanish\tnorms
dataset3\tIMAGEABILITY\timageability\tenglish\tratings
dataset4\tFREQUENCY_MEAN\tfrequency\tenglish\tnorms
dataset4\tFREQUENCY_LOG\tfrequency\tenglish\tnorms
dataset5\tDOMINANCE\tdominance\tenglish\tratings
dataset5\tEMOTIONALITY_MEAN\temotionality\tenglish\tnorms
dataset5\tEMOTIONALITY_MAX\temotionality\tenglish\tnorms
dataset6\tFAMILIARITY\tfamiliarity\tenglish\tratings
dataset6\tCONCRETENESS_MEAN\tconcreteness\tenglish\tratings
dataset6\tCONCRETENESS_SHORT\tconcreteness\tenglish\tratings
other_row\tOTHER_COL\tunknown_type\tenglish\tother
"""
    
    # Create zip file with TSV
    with zipfile.ZipFile(zip_path, 'w') as zf:
        zf.writestr("norare.tsv", tsv_content)
    
    return zip_path

zip_path = create_test_zip()

ARGS = {
    'zip_path': zip_path,
    'inner_tsv': 'norare.tsv'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
parse_language_tag Ran
ran; our contract draft was wrong, not the code
WPoelman/qwanqwa/src/qq/bcp47.py
code served (permissive licence) · get_code("636fc22b10df1422")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ParsedLanguageTag. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

from dataclasses import dataclass

@dataclass(frozen=True)
class ParsedLanguageTag:
    """Simple parsed result of a BCP 47 code.

    This intentionally does not model the full BCP 47 parts, only the
    primary language subtag and script/region pieces such as ``sr-Latn-RS`` and ``deu_Latn``.
    """
    original: str
    normalized: str
    language: str | None
    script: str | None = None
    region: str | None = None

def _is_region_subtag(subtag: str) -> bool:
    return len(subtag) == 2 and subtag.isalpha() or (len(subtag) == 3 and subtag.isdigit())

def parse_language_tag(tag: str) -> ParsedLanguageTag:
    """Parse a BCP-47-like tag and return its language/script/region pieces.

    The parser is deliberately permissive:
    - accepts ``-`` and ``_`` separators;
    - normalizes language/script/region casing;
    - ignores extensions/private-use subtags;
    - returns ``language=None`` for private-use-only tags such as ``x-foo``.
    """
    original = tag
    parts = [p for p in tag.replace('_', '-').split('-') if p]
    if not parts:
        return ParsedLanguageTag(original=original, normalized='', language=None)
    language = parts[0].lower()
    if language == 'x':
        normalized = '-'.join(('x', *(p.lower() for p in parts[1:])))
        return ParsedLanguageTag(original=original, normalized=normalized, language=None)
    script: str | None = None
    region: str | None = None
    i = 1
    if i < len(parts) and len(parts[i]) == 4 and parts[i].isalpha():
        script = parts[i].title()
        i += 1
    if i < len(parts) and _is_region_subtag(parts[i]):
        region = parts[i].upper()
        i += 1
    normalized_parts = [language]
    if script:
        normalized_parts.append(script)
    if region:
        normalized_parts.append(region)
    return ParsedLanguageTag(original=original, normalized='-'.join(normalized_parts), language=language, script=script, region=region)


ARGS = {
    "tag": "sr-Latn-RS"
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ParsedLanguageTag — a ParsedLanguageTag output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a ParsedLanguageTag output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
script Ran
ran; our contract draft was wrong, not the code
WPoelman/qwanqwa/case-studies/latex-tables/generate.py
code served (permissive licence) · get_code("d090146fe9795fc9")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

class MockScript:
    def __init__(self, iso_15924):
        self.iso_15924 = iso_15924

class MockLanguage:
    def __init__(self, canonical_scripts=None, script_codes=None):
        self.canonical_scripts = canonical_scripts if canonical_scripts is not None else []
        self.script_codes = script_codes if script_codes is not None else []

# Test case 1: language with canonical_scripts
lang_with_canonical = MockLanguage(
    canonical_scripts=[MockScript('Latn')],
    script_codes=['Zyyy']
)

ARGS = {
    'lang': lang_with_canonical
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
speakers Ran
ran; our contract draft was wrong, not the code
WPoelman/qwanqwa/case-studies/latex-tables/generate.py
code served (permissive licence) · get_code("299f41d69a5fec21")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

# The function 'speakers' takes an optional integer n and returns a formatted string representation
# It doesn't use any complex objects, just needs an integer or None

ARGS = {
    'n': 1500000
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
tex Ran
ran; our contract draft was wrong, not the code
WPoelman/qwanqwa/case-studies/latex-tables/generate.py
code served (permissive licence) · get_code("cac6fc4f14bc78f6")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

def tex(s: str | None) -> str:
    if not s:
        return '---'
    return s.replace('&', '\\&').replace('%', '\\%').replace('$', '\\$').replace('#', '\\#').replace('_', '\\_')

ARGS = {
    's': 'Hello & World % test $ with # special _ chars'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
tint Ran
ran, honoured the contract we drafted
WPoelman/qwanqwa/case-studies/huggingface-audit/analyze.py
code served (permissive licence) · get_code("74e807948c899ee4")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

# Analyzing the function:
# - color: str parameter that should be a valid hex color (e.g., "#RRGGBB")
# - white_mix: float parameter with default 0.18
# The function extracts RGB values from hex string and mixes with white

ARGS = {
    "color": "#FF5733",
    "white_mix": 0.18
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 1, dtype family float — a tuple output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
tint Ran
ran, honoured the contract we drafted
WPoelman/qwanqwa/case-studies/identifier-coverage/plot.py
code served (permissive licence) · get_code("3a40af0b6b12cc5b")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import matplotlib as mpl

ARGS = {
    "color": "blue",
    "white_mix": 0.18
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
matplotlib — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 1, dtype family float — a tuple output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
_tuple_distance Not yet run
unverified (harvested, no recorded run)
rspeer/langcodes/langcodes/language_distance.py
code served (permissive licence) · get_code("54fc5f9ea1714cfd")
How it ran

no invocation recorded.

fetch_hf_metadata Not yet run
unverified (harvested, no recorded run)
WPoelman/qwanqwa/case-studies/huggingface-audit/analyze.py
code served (permissive licence) · get_code("8fc94f22a36215f6")
How it ran

no invocation recorded.

Also present in other papers' code

Of 19 samples lifted from this paper's linked repositories, 19 carry a function identity and 0 do not. Of the 19 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 19 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Multilingual NLP research increasingly involves hundreds or thousands of languages across different datasets. Managing, discovering, and reporting language metadata becomes a common hurdle at these scales. We present QQ, a metadata toolkit and browser explorer. QQ compiles language metadata sources into a graph of language varieties, scripts, regions, identifiers, names, and relations, and exposes it through a Python API, a command-line interface, and a browser-based explorer. Users can normalize identifiers, retrieve metadata, traverse relations, and discover which external resources contain a language. We demonstrate QQ on three workflows: an audit of the Hug-gingFace Hub, linking resources that use different identifier systems, and generating reproducible language-reporting tables. QQ supports FAIR-oriented metadata practices through versioning, open formats, and reusable interfaces.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2603.00620")
get_code_for_paper("2603.00620")
have("2603.00620")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2603.00620.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2603.00620.svg)](https://syntology.ai/paper/2603.00620)

Connect an agent — have() is free.