SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2210.01936 · ICLR · 2023

When and why vision-language models behave like bags-of-words, and what to do about it?

James Zou, Dan Jurafsky, Mert Yuksekgonul, Pratyusha Kalluri, Federico Bianchi

arXiv · PDF · Open in the Atlas

Code that ran

3 of 8 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, 0 violated it and 0 have no contract verdict; 3 failures of our instrument; 5 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/2210.01936.json · badge: /api/ran/2210.01936.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
mertyg/vision-language-models-are-bows canonical (classifier label)
classifier basis not recorded on the link
3 ran · 5 no recorded run
FunctionStatusWhere it lives
get_image_perturb_fn Ran
ran; our contract draft was wrong, not the code
mertyg/vision-language-models-are-bows/dataset_zoo/perturbations.py
pointer only (licence: MIT); the stored copy get_code serves, from amitakamath/hard_positives, is pointer only (licence: NONE) · get_code("d52f072444282bb2")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from amitakamath/hard_positives, and GitHub detected no licence in that repository 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.

pre_caption Ran
ran; our contract draft was wrong, not the code
mertyg/vision-language-models-are-bows/dataset_zoo/retrieval.py
pointer only (licence: MIT); the stored copy get_code serves, from amitakamath/hard_positives, is pointer only (licence: NONE) · get_code("0d760ca65749dc14")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from amitakamath/hard_positives, and GitHub detected no licence in that repository 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.

shuffle_rows Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
mertyg/vision-language-models-are-bows/dataset_zoo/perturbations.py
pointer only (licence: MIT); the stored copy get_code serves, from amitakamath/hard_positives, is pointer only (licence: NONE) · get_code("29d83052b338a1f7")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from amitakamath/hard_positives, and GitHub detected no licence in that repository 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.

blip_decoder Not yet run
unverified (harvested, no recorded run)
mertyg/vision-language-models-are-bows/model_zoo/blip_utils/blip.py
pointer only (licence: MIT); the stored copy get_code serves, from wjpoom/spec, is pointer only (licence: NONE) · get_code("1bd954c1ba107cb9")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from wjpoom/spec, and GitHub detected no licence in that repository 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. No invocation recorded.

blip_feature_extractor Not yet run
unverified (harvested, no recorded run)
mertyg/vision-language-models-are-bows/model_zoo/blip_utils/blip.py
pointer only (licence: MIT); the stored copy get_code serves, from wjpoom/spec, is pointer only (licence: NONE) · get_code("16c013eb492b2d8c")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from wjpoom/spec, and GitHub detected no licence in that repository 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. No invocation recorded.

blip_itm Not yet run
unverified (harvested, no recorded run)
mertyg/vision-language-models-are-bows/model_zoo/blip_utils/blip_itm.py
pointer only (licence: MIT); the stored copy get_code serves, from wjpoom/spec, is pointer only (licence: NONE) · get_code("84d0b6c25badc1c3")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from wjpoom/spec, and GitHub detected no licence in that repository 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. No invocation recorded.

create_vit Not yet run
unverified (harvested, no recorded run)
mertyg/vision-language-models-are-bows/model_zoo/blip_utils/blip.py
code served (permissive licence) · get_code("f837c334ddfa8af4")
How it ran

no invocation recorded.

get_text_perturb_fn Not yet run
unverified (harvested, no recorded run)
mertyg/vision-language-models-are-bows/dataset_zoo/perturbations.py
pointer only (licence: MIT); the stored copy get_code serves, from amitakamath/hard_positives, is pointer only (licence: NONE) · get_code("5c8ae0e86a1836f6")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from amitakamath/hard_positives, and GitHub detected no licence in that repository 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. No invocation recorded.

Also present in other papers' code

Of 8 samples lifted from this paper's linked repositories, 8 carry a function identity and 0 do not. Of the 8 distinct functions with an identity, 8 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 8 later) and 0 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.

FunctionAlso present inEarliest-datedOthers named
create_vit 5 other papers
5 later
2304.05977 (2023-04)
ImageReward: Learning and Evaluating Human Preferences for Text-to-Image Generation
2305.16304 (2023-05)
Candidate Set Re-ranking for Composed Image Retrieval with Dual Multi-modal Encoder
2310.19785 (2023-10)
What's "up" with vision-language models? Investigating their struggle with spatial reasoning
and 2 more not named here
pre_caption 5 other papers
5 later
2303.10323 (2023-03)
Dynamic Graph Enhanced Contrastive Learning for Chest X-ray Report Generation
2310.07972 (2023-10)
Interpretable Diffusion via Information Decomposition
2310.19785 (2023-10)
What's "up" with vision-language models? Investigating their struggle with spatial reasoning
and 2 more not named here
blip_decoder 2 other papers
2 later
2310.19785 (2023-10)
What's "up" with vision-language models? Investigating their struggle with spatial reasoning
2312.00081 (2023-12)
Synthesize, Diagnose, and Optimize: Towards Fine-Grained Vision-Language Understanding
blip_feature_extractor 2 other papers
2 later
2310.19785 (2023-10)
What's "up" with vision-language models? Investigating their struggle with spatial reasoning
2312.00081 (2023-12)
Synthesize, Diagnose, and Optimize: Towards Fine-Grained Vision-Language Understanding
get_image_perturb_fn 2 other papers
2 later
2310.19785 (2023-10)
What's "up" with vision-language models? Investigating their struggle with spatial reasoning
2409.17958 (2024-09)
The Hard Positive Truth about Vision-Language Compositionality
get_text_perturb_fn 2 other papers
2 later
2310.19785 (2023-10)
What's "up" with vision-language models? Investigating their struggle with spatial reasoning
2409.17958 (2024-09)
The Hard Positive Truth about Vision-Language Compositionality
shuffle_rows 2 other papers
2 later
2310.19785 (2023-10)
What's "up" with vision-language models? Investigating their struggle with spatial reasoning
2409.17958 (2024-09)
The Hard Positive Truth about Vision-Language Compositionality
blip_itm 1 other paper
1 later
2312.00081 (2023-12)
Synthesize, Diagnose, and Optimize: Towards Fine-Grained Vision-Language Understanding

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

Despite the use of large vision and language models (VLMs) in many downstream applications, it is unclear how well they encode the compositional relationships between objects and attributes. Here, we create the Attribution, Relation, and Order (ARO) benchmark to systematically evaluate the ability of VLMs to understand different types of relationships, attributes, and order information. ARO consists of Visual Genome Attribution, to test the understanding of objects' properties; Visual Genome Relation, to test for relational understanding; and COCO-Order & Flickr30k-Order, to test for order sensitivity in VLMs. ARO is orders of magnitude larger than previous benchmarks of compositionality, with more than 50,000 test cases. We present the settings in which state-of-the-art VLMs behave like bagsof-words-i.e. when they have poor relational understanding, can blunder when linking objects to their attributes, and demonstrate a severe lack of order sensitivity. VLMs are predominantly trained and evaluated on large scale datasets with rich compositional structure in the images and captions. Yet, training on these datasets has not been enough to address the lack of compositional understanding, and evaluating on these datasets has failed to surface this deficiency. To understand why these limitations emerge and are not represented in the standard tests, we zoom into the training and evaluation procedures. We demonstrate that it is possible to perform well on image-text retrieval over existing datasets without using the composition and order information. This further motivates the value of using ARO to benchmark VLMs. Given that contrastive pretraining optimizes for retrieval on large datasets with similar shortcuts, we hypothesize that this can explain why the models do not need to learn to represent compositional information. This finding suggests a natural solution: composition-aware hard negative mining. We show that a simple-to-implement modification of contrastive learning significantly improves the performance on tasks requiring an understanding of order and compositionality.

For agents

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

get_harvested_code_for_paper("2210.01936")
get_code_for_paper("2210.01936")
have("2210.01936")

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

curl https://syntology.ai/api/ran/2210.01936.json

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

[![Syntology run record](https://syntology.ai/api/ran/2210.01936.svg)](https://syntology.ai/paper/2210.01936)

Connect an agent — have() is free.