SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2305.05976 · ACL · 2023

Say What You Mean! Large Language Models Speak Too Positively about Negative Commonsense Knowledge

Lei Li, Sijie Cheng, Yanghua Xiao, Jiangjie Chen, Wei Shi, Ziquan Fu, / Examples

arXiv · PDF · Open in the Atlas

Code that ran

4 of 4 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; 1 honoured the contract we drafted, 1 violated it and 0 have no contract verdict; 2 failures of our instrument; 0 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/2305.05976.json · badge: /api/ran/2305.05976.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
jiangjiechen/uncommongen canonical (classifier label)
classifier basis: classifier_2025
4 ran · 0 no recorded run
FunctionStatusWhere it lives
detect_negation Ran
ran, violated the contract we drafted
jiangjiechen/uncommongen/evaluation/eval_constrained_generation.py
code served (permissive licence) · get_code("dc0fd60baf11ab5b")
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) 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 re

negation_cue = {'nothing', 'no', 'not', 'never', 'none', 'hardly', 'rarely', 'scarcely', 'seldom', 'barely', 'nor', 'neither', 'nothing', 'nowhere', 'without', 'lack', 'cant', 'dont', 'doesnt', "doesn't", "don't", "isn't", "wasn't", "aren't", "weren't", "haven't", "hasn't", "shouldn't", "won't", "wouldn't", "can't", "couldn't", 'cannot', 'unable'}

def fetch_target_sent(x, prefix):
    if prefix is not None:
        for subx in x.strip().replace('###', '\n').split('\n'):
            sent = re.findall(f'{prefix}(.*)', subx)
            if sent != []:
                return sent[0].strip()
    return x.strip().split('\n')[0].strip()

def detect_negation(pipe, texts: list, prefix=None, only_rule=False):

    def _has_neg(x):
        for j in x:
            if j['entity'] == 'Y':
                return True
        return False
    labels = []
    for t in texts:
        rule_hit_flag = False
        t1 = fetch_target_sent(t, prefix)
        if only_rule:
            label = False
            for w in t1.split():
                if w.lower() in negation_cue:
                    label = True
                    break
            labels.append(label)
        else:
            for w in t1.split():
                if w.lower() in negation_cue:
                    labels.append(True)
                    rule_hit_flag = True
                    break
            if not rule_hit_flag:
                labels.append(_has_neg(pipe(t1)))
    return labels


class MockPipe:
    """Mock NER pipeline that returns entity predictions."""
    def __call__(self, text):
        # Return a list of entity dicts with 'entity' key
        # Simulate some entities, some with 'Y' label
        return [
            {'entity': 'N', 'word': 'test'},
            {'entity': 'Y', 'word': 'label'}
        ]


# Create mock pipe
mock_pipe = MockPipe()

# Create test texts
test_texts = [
    "This is a sentence with no negation",
    "This sentence has not any value",
    "Another text without issues"
]

ARGS = {
    'pipe': mock_pipe,
    'texts': test_texts,
    'prefix': None,
    'only_rule': False
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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 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
fetch_target_sent Ran
ran; our contract draft was wrong, not the code
jiangjiechen/uncommongen/evaluation/eval_constrained_generation.py
code served (permissive licence) · get_code("3ec060bf22be70f4")
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) 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 re

def fetch_target_sent(x, prefix):
    if prefix is not None:
        for subx in x.strip().replace('###', '\n').split('\n'):
            sent = re.findall(f'{prefix}(.*)', subx)
            if sent != []:
                return sent[0].strip()
    return x.strip().split('\n')[0].strip()

# Setup code
ARGS = {
    'x': 'PREFIX: hello world\n###\nPREFIX: second line',
    'prefix': 'PREFIX: '
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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
judge_ambi_w_prob Ran
ran; our contract draft was wrong, not the code
jiangjiechen/uncommongen/boolqa/llm_answer_prediction.py
code served (permissive licence) · get_code("9500a08d95e21455")
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) 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 torch.nn as nn

torch.manual_seed(0)

class MockPromptFunc:
    """Mock function that simulates the prompt_func behavior"""
    def __call__(self, prompts, echo=True, clean=False, logprobs=1, **kwargs):
        # Return mock results with logprobs for each prompt
        results = [
            {
                'logprobs': {
                    'token_logprobs': [-0.5, -0.3, 0.2]  # Last element is -0.3 for 'yes'
                }
            },
            {
                'logprobs': {
                    'token_logprobs': [-0.5, -0.4, -0.8]  # Last element is -0.8 for 'no'
                }
            }
        ]
        return results, 100  # Return results and mock money value

prompt = "Is this ambiguous?"
prompt_func = MockPromptFunc()

ARGS = {
    'prompt': prompt,
    'prompt_func': prompt_func,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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
majority_vote Ran
ran, honoured the contract we drafted
jiangjiechen/uncommongen/evaluation/eval_constrained_generation.py
code served (permissive licence) · get_code("379f828142e3c658")
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) 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.

from collections import defaultdict, Counter

# majority_vote takes pred_list as input
# It iterates over x in pred_list, and for each x it calls Counter(x).most_common()[0][0]
# This means:
# - pred_list is an iterable of iterables
# - Each element x should be countable (list, tuple, etc.)
# - Counter(x) creates a counter from x
# - .most_common() returns list of (element, count) tuples sorted by count
# - [0][0] gets the most common element

# Create a simple pred_list with multiple predictions
# Each inner list represents predictions from different models for a single sample
pred_list = [
    [1, 1, 2, 1],      # Sample 1: predictions are [1, 1, 2, 1], majority is 1
    [0, 0, 1, 0],      # Sample 2: predictions are [0, 0, 1, 0], majority is 0
    ['a', 'b', 'a', 'a'],  # Sample 3: predictions are ['a', 'b', 'a', 'a'], majority is 'a'
]

ARGS = {
    'pred_list': pred_list
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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 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

Also present in other papers' code

Of 4 samples lifted from this paper's linked repositories, 4 carry a function identity and 0 do not. Of the 4 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 4 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

Large language models (LLMs) have been widely studied for their ability to store and utilize positive knowledge. However, negative knowledge, such as "lions don't live in the ocean", is also ubiquitous in the world but rarely mentioned explicitly in the text. What do LLMs know about negative knowledge? This work examines the ability of LLMs to negative commonsense knowledge. We design a constrained keywords-to-sentence generation task (CG) and a Boolean question-answering task (QA) to probe LLMs. Our experiments reveal that LLMs frequently fail to generate valid sentences grounded in negative commonsense knowledge, yet they can correctly answer polar yes-or-no questions. We term this phenomenon the belief conflict of LLMs. Our further analysis shows that statistical shortcuts and negation reporting bias from language modeling pre-training cause this conflict. 1

For agents

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

get_harvested_code_for_paper("2305.05976")
get_code_for_paper("2305.05976")
have("2305.05976")

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

curl https://syntology.ai/api/ran/2305.05976.json

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

[![Syntology run record](https://syntology.ai/api/ran/2305.05976.svg)](https://syntology.ai/paper/2305.05976)

Connect an agent — have() is free.