SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 1909.06092 · 2019

A General Framework for Implicit and Explicit Debiasing of Distributional Word Vector Spaces

arXiv · PDF · Open in the Atlas

Code that ran

18 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; 0 honoured the contract we drafted, 1 violated it and 17 have no contract verdict; 0 failures of our instrument; 1 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; 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/1909.06092.json · badge: /api/ran/1909.06092.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
nfriedri/debie-backend community (classifier label)
classifier basis not recorded on the link
14 ran · 0 no recorded run
umanlp/debie-command-line community (classifier label)
classifier basis not recorded on the link
3 ran · 1 no recorded run
anlausch/debie reimplementation (classifier label)
classifier basis: classifier_2025
1 ran · 0 no recorded run
FunctionStatusWhere it lives
bias_analogy_test Ran
ran on a synthesized input
umanlp/debie-command-line/bias_evaluation/bat.py
code served (permissive licence) · get_code("9b616b12c15d0fcb")
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 float, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape 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 numpy as np
import logging

logging.basicConfig(level=logging.INFO)

# Create small embedding vectors (dimension 4)
embedding_dim = 4

# target1: dict of words -> vector representations
target1 = {
    "word1": np.random.randn(embedding_dim),
    "word2": np.random.randn(embedding_dim),
}

# target2: dict of words -> vector representations
target2 = {
    "word3": np.random.randn(embedding_dim),
    "word4": np.random.randn(embedding_dim),
}

# attribute1: dict of attribute words -> vector representations
attribute1 = {
    "attr1": np.random.randn(embedding_dim),
    "attr2": np.random.randn(embedding_dim),
}

# attribute2: dict of attribute words -> vector representations
attribute2 = {
    "attr3": np.random.randn(embedding_dim),
    "attr4": np.random.randn(embedding_dim),
}

ARGS = {
    "target1": target1,
    "target2": target2,
    "attribute1": attribute1,
    "attribute2": attribute2,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type float, ndim 0, dtype family float, shape () — shape () as a later run recorded it
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
boolean_string Ran
ran, violated the contract we drafted
anlausch/debie/exp_debiasnet_args.py
pointer only (licence: NONE) · get_code("d481e95ca1390889")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, anlausch/debie when it was checked. 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.

create_duplicates Ran
ran on a synthesized input
nfriedri/debie-backend/calculation.py
code served (permissive licence) · get_code("559fa36b836d0d46")
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. 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 copy
import torch
import numpy as np

torch.manual_seed(0)

# The function uses copy.deepcopy on the arguments, so we need objects that can be deep copied
# Let's create various types of objects that are commonly used

# Simple tensor
set1_value = torch.randn(2, 4)

# Dictionary
set2_value = {
    "weights": torch.randn(3, 4),
    "bias": torch.randn(4),
    "metadata": {"name": "layer1", "count": 5}
}

# List of mixed types
set3_value = [
    torch.randn(2, 4),
    np.array([1, 2, 3]),
    {"key": "value"},
    42
]

# Nested structure
set4_value = {
    "tensors": [torch.randn(2, 3), torch.randn(3, 5)],
    "arrays": np.array([[1, 2], [3, 4]]),
    "nested": {
        "inner_tensor": torch.randn(4, 2),
        "inner_list": [1, 2, 3]
    }
}

ARGS = {
    "set1": set1_value,
    "set2": set2_value,
    "set3": set3_value,
    "set4": set4_value
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is 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
dict_keys_to_string Ran
ran on a synthesized input
nfriedri/debie-backend/json_controller.py
code served (permissive licence) · get_code("99bdfa1c386e0cc2")
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
import numpy as np

torch.manual_seed(0)

# The function iterates over vector_dict.keys() and converts each key to string
# It concatenates them with spaces, then removes the trailing space
# So vector_dict needs to be a dict with string-convertible keys

vector_dict = {
    'embedding': torch.randn(2, 4),
    'attention': torch.randn(3, 5),
    'output': torch.randn(4, 6)
}

ARGS = {
    'vector_dict': vector_dict
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, 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
dict_to_json Ran
ran on a synthesized input
nfriedri/debie-backend/json_controller.py
code served (permissive licence) · get_code("c21d9e8dd31a20a5")
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 torch
import numpy as np

torch.manual_seed(0)

# The function iterates over keys in vector_dict and calls .tolist() on each value
# So vector_dict should be a dict mapping strings to torch tensors (or numpy arrays)

vector_dict = {
    "word1": torch.randn(2, 4),
    "word2": torch.randn(3, 5),
    "word3": torch.randn(2, 6),
}

ARGS = {
    "vector_dict": vector_dict
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — 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
eval_svm Ran
ran on a synthesized input
umanlp/debie-command-line/bias_evaluation/svm_classifier.py
code served (permissive licence) · get_code("83515675d0094fcb")
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 float, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape 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 logging
import random
import numpy as np
from sklearn import svm

logging.basicConfig(level=logging.INFO)

# Create a small vocabulary
vocab = {
    'word1': 0,
    'word2': 1,
    'word3': 2,
    'word4': 3,
    'word5': 4,
    'word6': 5,
    'word7': 6,
    'word8': 7,
}

# Create word vectors (embedding vectors for each word in vocab)
# Shape: (num_words, embedding_dim) = (8, 4)
vecs = np.random.randn(8, 4).astype(np.float32)

# Create augmentations (lists of words from the vocabulary)
augmentations1 = ['word1', 'word2', 'word3', 'word4']
augmentations2 = ['word5', 'word6', 'word7', 'word8']

# Create target sets (lists of words from the vocabulary)
target1 = ['word1', 'word2']
target2 = ['word5', 'word6']

ARGS = {
    'augmentations1': augmentations1,
    'augmentations2': augmentations2,
    'target1': target1,
    'target2': target2,
    'vocab': vocab,
    'vecs': vecs,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, sklearn — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type float, ndim 0, dtype family float, shape () — shape () as a later run recorded it
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_set_sizes Ran
ran on a synthesized input
nfriedri/debie-backend/specification_controller.py
code served (permissive licence) · get_code("ce02a6d9b77ea470")
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. 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 logging
import random

logging.basicConfig(level=logging.INFO)

# The function expects two dictionaries that support:
# - len() operation
# - .keys() method
# - del operation
# - random.choice() on keys

# Create two dictionaries of different sizes
vector_set1 = {
    'key_0': [1, 2, 3, 4],
    'key_1': [5, 6, 7, 8],
    'key_2': [9, 10, 11, 12],
    'key_3': [13, 14, 15, 16],
    'key_4': [17, 18, 19, 20],
}

vector_set2 = {
    'vec_0': [21, 22, 23, 24],
    'vec_1': [25, 26, 27, 28],
}

ARGS = {
    'vector_set1': vector_set1,
    'vector_set2': vector_set2,
}
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 — a tuple output has no array shape of its own; none is 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
format_sets Ran
ran on a synthesized input
nfriedri/debie-backend/specification_controller.py
code served (permissive licence) · get_code("33edd73c1410848f")
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. 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 logging
import random
import torch

logging.basicConfig(level=logging.INFO)

# Set seed for reproducibility
torch.manual_seed(0)
random.seed(0)

# Create test dictionaries with different sizes
# vector_set1 will have 5 items
vector_set1_t = {
    'key_0': torch.randn(4, 3),
    'key_1': torch.randn(4, 3),
    'key_2': torch.randn(4, 3),
    'key_3': torch.randn(4, 3),
    'key_4': torch.randn(4, 3),
}

# vector_set2 will have 3 items
vector_set2_t = {
    'key_5': torch.randn(4, 3),
    'key_6': torch.randn(4, 3),
    'key_7': torch.randn(4, 3),
}

# vector_set1 will have 2 items
vector_set1_a = {
    'key_8': torch.randn(4, 3),
    'key_9': torch.randn(4, 3),
}

# vector_set2 will have 4 items
vector_set2_a = {
    'key_10': torch.randn(4, 3),
    'key_11': torch.randn(4, 3),
    'key_12': torch.randn(4, 3),
    'key_13': torch.randn(4, 3),
}

ARGS = {
    't1': vector_set1_t,
    't2': vector_set2_t,
    'a1': vector_set1_a,
    'a2': vector_set2_a,
}
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 — a tuple output has no array shape of its own; none is 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
get_vector_from_small_dict Ran
ran on a synthesized input
nfriedri/debie-backend/Version0.1/compute_analogy.py
code served (permissive licence) · get_code("bcfffd82f19b8867")
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 list, 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

torch.manual_seed(0)

dict1 = {
    "word1": [1.0, 2.0, 3.0, 4.0],
    "word2": [5.0, 6.0, 7.0, 8.0],
    "word3": [9.0, 10.0, 11.0, 12.0]
}

ARGS = {
    "dict1": dict1
}
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 list, ndim 1, dtype family float — a list 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
json_vector_retrieval Ran
ran on a synthesized input
nfriedri/debie-backend/json_controller.py
code served (permissive licence) · get_code("0e30b32211028bc2")
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
import numpy as np

torch.manual_seed(0)

# The function json_vector_retrieval expects:
# 1. vector_dict: a dict where keys are words and values are numpy arrays or torch tensors with .tolist() method
# 2. not_found: a value that can be converted to string

# Create sample vector_dict with torch tensors
vector_dict = {
    'hello': torch.tensor([1.0, 2.0, 3.0, 4.0]),
    'world': torch.tensor([5.0, 6.0, 7.0, 8.0]),
    'python': torch.tensor([9.0, 10.0, 11.0, 12.0])
}

# Create not_found list
not_found = ['unknown', 'missing', 'notfound']

ARGS = {
    'vector_dict': vector_dict,
    'not_found': not_found
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, 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
load_binary_embeddings Ran
ran on a synthesized input
nfriedri/debie-backend/data_controller.py
code served (permissive licence) · get_code("11c6271f3ac7fad6")
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. 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 pickle
import numpy as np
import tempfile
import os

# Create temporary directory for test files
temp_dir = tempfile.mkdtemp()

# Create a small vocab pickle file
vocab = {'hello': 0, 'world': 1, 'test': 2, 'python': 3}
vocab_path = os.path.join(temp_dir, 'vocab.pkl')
with open(vocab_path, 'wb') as f:
    pickle.dump(vocab, f)

# Create a small vectors numpy file
vecs = np.array([
    [0.1, 0.2, 0.3, 0.4],
    [0.5, 0.6, 0.7, 0.8],
    [0.2, 0.3, 0.4, 0.5],
    [0.9, 0.1, 0.2, 0.3]
], dtype=np.float32)
vecs_path = os.path.join(temp_dir, 'vecs.npy')
np.save(vecs_path, vecs)

# Define arguments for load_binary_embeddings
ARGS = {
    'vocab_path': vocab_path,
    'vecs_path': vecs_path,
    'inverse': False,
    'normalize': False
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is 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_vectors_binary Ran
ran on a synthesized input
nfriedri/debie-backend/data_controller.py
code served (permissive licence) · get_code("afd00edba761da8c")
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 ndarray, ndim 2, dtype family float, shape (5, 8) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape 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 numpy as np
import tempfile
import os

# Create a temporary directory and file for the test
temp_dir = tempfile.mkdtemp()
temp_file = os.path.join(temp_dir, "vectors.npy")

# Create a small sample numpy array and save it
sample_vecs = np.random.randn(5, 8).astype(np.float32)
np.save(temp_file, sample_vecs)

ARGS = {
    "path": temp_file,
    "normalize": False
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family float, shape (5, 8) — shape (5, 8) as a later run recorded it
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_vocab_binary Ran
ran on a synthesized input
nfriedri/debie-backend/data_controller.py
code served (permissive licence) · get_code("0f45921d6c72d3da")
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 pickle
import tempfile
import os

# Create a temporary directory for test files
temp_dir = tempfile.mkdtemp()

# Create a sample vocabulary dictionary
sample_vocab = {
    'hello': 0,
    'world': 1,
    'foo': 2,
    'bar': 3,
    'baz': 4
}

# Save the vocabulary to a pickle file
vocab_path = os.path.join(temp_dir, 'vocab.pkl')
with open(vocab_path, 'wb') as f:
    pickle.dump(sample_vocab, f)

# Define arguments for the function
ARGS = {
    'path': vocab_path,
    'inverse': False
}
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
retrieve_vector_multiple Ran
ran on a synthesized input
nfriedri/debie-backend/vector_retrieval.py
code served (permissive licence) · get_code("f5d190ff20783914")
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. 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
import numpy as np

torch.manual_seed(0)

# vocab: dict mapping words to indices
vocab = {
    "hello": 0,
    "world": 1,
    "python": 2,
    "test": 3
}

# vectors: tensor or array-like where vectors[i] gives the vector for word at index i
vectors = torch.randn(4, 4)  # 4 words, 4-dimensional vectors

# word_list: list of words to retrieve vectors for
word_list = ["hello", "world", "unknown", "python"]

ARGS = {
    "vocab": vocab,
    "vectors": vectors,
    "word_list": word_list
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is 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
retrieve_vector_single Ran
ran on a synthesized input
nfriedri/debie-backend/vector_retrieval.py
code served (permissive licence) · get_code("261fa31e43f56cbc")
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. 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
import numpy as np

torch.manual_seed(0)

# vocab: dict mapping words (str) to indices (int)
vocab = {
    "hello": 0,
    "world": 1,
    "python": 2
}

# vectors: tensor or array where vectors[i] gives the embedding for vocab word at index i
# Shape: (vocab_size, embedding_dim)
vectors = torch.randn(3, 4)

# target_word: str to look up
target_word = "hello"

ARGS = {
    "vocab": vocab,
    "vectors": vectors,
    "target_word": target_word
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is 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
target_set_mean_vector Ran
ran on a synthesized input
umanlp/debie-command-line/bias_evaluation/ect.py
code served (permissive licence) · get_code("549b1aad49e944c5")
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 ndarray, ndim 1, dtype family float, shape (300,) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape 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 numpy

torch_manual_seed = 0

target_list = [
    numpy.ones(300),
    numpy.ones(300) * 2,
    numpy.ones(300) * 3,
]

ARGS = {
    "target_list": target_list,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 1, dtype family float, shape (300,) — shape (300,) as a later run recorded it
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
transform_dict_to_list Ran
ran on a synthesized input
nfriedri/debie-backend/calculation.py
code served (permissive licence) · get_code("ccdaa630046f3a52")
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 ndarray, ndim 2, dtype family float, shape (3, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape 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 numpy
import torch

# The function takes a dictionary where values are array-like objects
# and converts them to a list of lists, then to a numpy array

# Create a simple dict with word keys and array-like values
dict1 = {
    'word1': [1.0, 2.0, 3.0, 4.0],
    'word2': [5.0, 6.0, 7.0, 8.0],
    'word3': [9.0, 10.0, 11.0, 12.0]
}

ARGS = {
    'dict1': dict1
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family float, shape (3, 4) — shape (3, 4) as a later run recorded it
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
transform_multiple_dicts_to_lists Ran
ran on a synthesized input
nfriedri/debie-backend/calculation.py
code served (permissive licence) · get_code("3e20da2273d79713")
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 3, 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 numpy

# Create small dictionaries with word -> vector mappings
dict1 = {
    "hello": [1.0, 2.0, 3.0, 4.0],
    "world": [5.0, 6.0, 7.0, 8.0]
}

dict2 = {
    "foo": [9.0, 10.0, 11.0, 12.0],
    "bar": [13.0, 14.0, 15.0, 16.0]
}

dict3 = {
    "alpha": [17.0, 18.0, 19.0, 20.0],
    "beta": [21.0, 22.0, 23.0, 24.0]
}

dict4 = {
    "one": [25.0, 26.0, 27.0, 28.0],
    "two": [29.0, 30.0, 31.0, 32.0]
}

ARGS = {
    "dict1": dict1,
    "dict2": dict2,
    "dict3": dict3,
    "dict4": dict4
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 3, 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_dict_uploaded_file Not yet run
unverified (harvested, no recorded run)
umanlp/debie-command-line/upload_controller.py
code served (permissive licence) · get_code("cfc69051efee0623")
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

Distributional word vectors have recently been shown to encode many of the human biases, most notably gender and racial biases, and models for attenuating such biases have consequently been proposed. However, existing models and studies (1) operate on under-specified and mutually differing bias definitions, (2) are tailored for a particular bias (e.g., gender bias) and (3) have been evaluated inconsistently and non-rigorously. In this work, we introduce a general framework for debiasing word embeddings. We operationalize the definition of a bias by discerning two types of bias specification: explicit and implicit. We then propose three debiasing models that operate on explicit or implicit bias specifications and that can be composed towards more robust debiasing. Finally, we devise a full-fledged evaluation framework in which we couple existing bias metrics with newly proposed ones. Experimental findings across three embedding methods suggest that the proposed debiasing models are robust and widely applicable: they often completely remove the bias both implicitly and explicitly without degradation of semantic information encoded in any of the input distributional spaces. Moreover, we successfully transfer debiasing models, by means of cross-lingual embedding spaces, and remove or attenuate biases in distributional word vector spaces of languages that lack readily available bias specifications.

For agents

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

get_harvested_code_for_paper("1909.06092")
get_code_for_paper("1909.06092")
have("1909.06092")

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

curl https://syntology.ai/api/ran/1909.06092.json

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

[![Syntology run record](https://syntology.ai/api/ran/1909.06092.svg)](https://syntology.ai/paper/1909.06092)

Connect an agent — have() is free.