FMAT

😷 The Fill-Mask Association Test (掩码填空联系测验).

The Fill-Mask Association Test (FMAT) is an integrative and probability-based method using BERT Models to measure conceptual associations (e.g., attitudes, biases, stereotypes, social norms, cultural values) as propositions in natural language (Bao, 2024, JPSP).

⚠️ Please update this package to version ≥ 2025.4 for faster and more robust functionality.

Author

Bruce H. W. S. Bao 包寒吴霜

📬 baohws@foxmail.com

📋 psychbruce.github.io

Citation

(1) FMAT Package

Bao, H. W. S. (2023). FMAT: The Fill-Mask Association Test. https://doi.org/10.32614/CRAN.package.FMAT

(2) FMAT Research Articles - Methodology

Bao, H. W. S. (2024). The Fill-Mask Association Test (FMAT): Measuring propositions in natural language. Journal of Personality and Social Psychology, 127(3), 537–561. https://doi.org/10.1037/pspa0000396

(3) FMAT Research Articles - Application

Bao, H. W. S., & Gries, P. (2024). Intersectional race–gender stereotypes in natural language. British Journal of Social Psychology, 63(4), 1771–1786. https://doi.org/10.1111/bjso.12748
Bao, H. W. S., & Gries, P. (2025). Biases about Chinese people in English language use: Stereotypes, prejudice and discrimination. China Quarterly. https://doi.org/10.1017/S0305741025100532
Wang, Z., Xia, H., Bao, H. W. S., Jing, Y., & Gu, R. (2025). Artificial intelligence is stereotypically linked more with socially dominant groups in natural language. Advanced Science. https://doi.org/10.1002/advs.202508623

Installation

The R package FMAT and three Python packages (transformers, torch, huggingface-hub) all need to be installed.

(1) R Package

## Method 1: Install from CRAN
install.packages("FMAT")

## Method 2: Install from GitHub
install.packages("devtools")
devtools::install_github("psychbruce/FMAT", force=TRUE)

(2) Python Environment and Packages

Install Anaconda (a recommended package manager that automatically installs Python, its IDEs like Spyder, and a large list of common Python packages).

Specify the Anaconda’s Python interpreter in RStudio.

RStudio → Tools → Global/Project Options
→ Python → Select → Conda Environments
→ Choose “…/Anaconda3/python.exe”

Install specific versions of Python packages “transformers”, “torch”, and “huggingface-hub”.
(RStudio Terminal / Anaconda Prompt / Windows Command)

For CPU users:

pip install transformers==4.40.2 torch==2.2.1 huggingface-hub==0.20.3

For GPU (CUDA) users:

pip install transformers==4.40.2 huggingface-hub==0.20.3
pip install torch==2.2.1 --index-url https://download.pytorch.org/whl/cu121

To use some models (e.g., microsoft/deberta-v3-base), “You need to have sentencepiece installed to convert a slow tokenizer to a fast one”:

pip install sentencepiece

See Guidance for GPU Acceleration for installation guidance if you have an NVIDIA GPU device on your PC and want to use GPU to accelerate the pipeline.
According to the May 2024 releases, “transformers” ≥ 4.41 depends on “huggingface-hub” ≥ 0.23. The suggested versions of “transformers” (4.40.2) and “huggingface-hub” (0.20.3) ensure the console display of progress bars when downloading BERT models while keeping these packages as new as possible.
Proxy users may use the “global mode” (全局模式) to download models.
If you find the error HTTPSConnectionPool(host='huggingface.co', port=443), please try to (1) reinstall Anaconda so that some unknown issues may be fixed, or (2) downgrade the “urllib3” package to version ≤ 1.25.11 (pip install urllib3==1.25.11) so that it will use HTTP proxies (rather than HTTPS proxies as in later versions) to connect to Hugging Face.

Guidance for FMAT

Step 1: Download BERT Models

Use BERT_download() to download BERT models. Model files are saved in your local cache folder “%USERPROFILE%/.cache/huggingface”. A full list of BERT models are available at Hugging Face.

Use BERT_info() and BERT_vocab() to obtain detailed information of BERT models.

Step 2: Design FMAT Queries

Design queries that conceptually represent the constructs you would measure (see Bao, 2024, JPSP for how to design queries).

Use FMAT_query() and/or FMAT_query_bind() to prepare a data.table of queries.

Step 3: Run FMAT

Use FMAT_run() to get raw data (probability estimates) for further analysis.

Several steps of preprocessing have been included in the function for easier use (see FMAT_run() for details).

For BERT variants using <mask> rather than [MASK] as the mask token, the input query will be automatically modified so that users can always use [MASK] in query design.
For some BERT variants, special prefix characters such as \u0120 and \u2581 will be automatically added to match the whole words (rather than subwords) for [MASK].

Notes

Improvements are ongoing, especially for adaptation to more diverse (less popular) BERT models.
If you find bugs or have problems using the functions, please report them at GitHub Issues or send me an email.

Guidance for GPU Acceleration

By default, the FMAT package uses CPU to enable the functionality for all users. But for advanced users who want to accelerate the pipeline with GPU, the FMAT_run() function now supports using a GPU device, about 3x faster than CPU.

Test results (on the developer’s computer, depending on BERT model size):

CPU (Intel 13th-Gen i7-1355U): 500~1000 queries/min
GPU (NVIDIA GeForce RTX 2050): 1500~3000 queries/min

Checklist:

Ensure that you have an NVIDIA GPU device (e.g., GeForce RTX Series) and an NVIDIA GPU driver installed on your system.
Install PyTorch (Python torch package) with CUDA support.
- Find guidance for installation command at https://pytorch.org/get-started/locally/.
- CUDA is available only on Windows and Linux, but not on MacOS.
- If you have installed a version of torch without CUDA support, please first uninstall it (command: pip uninstall torch) and then install the suggested one.
- You may also install the corresponding version of CUDA Toolkit (e.g., for the torch version supporting CUDA 12.1, the same version of CUDA Toolkit 12.1 may also be installed).

Example code for installing PyTorch with CUDA support:
(RStudio Terminal / Anaconda Prompt / Windows Command)

pip install torch==2.2.1 --index-url https://download.pytorch.org/whl/cu121

BERT Models

The reliability and validity of the following 12 BERT models in the FMAT have been established in our research, but future work is needed to examine the performance of other models.

(model name on Hugging Face - model file size)

For details about BERT, see:

library(FMAT)
models = c(
  "bert-base-uncased",
  "bert-base-cased",
  "bert-large-uncased",
  "bert-large-cased",
  "distilbert-base-uncased",
  "distilbert-base-cased",
  "albert-base-v1",
  "albert-base-v2",
  "roberta-base",
  "distilroberta-base",
  "vinai/bertweet-base",
  "vinai/bertweet-large"
)
BERT_download(models)

ℹ Device Info:

R Packages:
FMAT          2024.5
reticulate    1.36.1

Python Packages:
transformers  4.40.2
torch         2.2.1+cu121

NVIDIA GPU CUDA Support:
CUDA Enabled: TRUE
CUDA Version: 12.1
GPU (Device): NVIDIA GeForce RTX 2050


── Downloading model "bert-base-uncased" ──────────────────────────────────────────
→ (1) Downloading configuration...
config.json: 100%|██████████| 570/570 [00:00<00:00, 114kB/s]
→ (2) Downloading tokenizer...
tokenizer_config.json: 100%|██████████| 48.0/48.0 [00:00<00:00, 23.9kB/s]
vocab.txt: 100%|██████████| 232k/232k [00:00<00:00, 1.50MB/s]
tokenizer.json: 100%|██████████| 466k/466k [00:00<00:00, 1.98MB/s]
→ (3) Downloading model...
model.safetensors: 100%|██████████| 440M/440M [00:36<00:00, 12.1MB/s] 
✔ Successfully downloaded model "bert-base-uncased"

── Downloading model "bert-base-cased" ────────────────────────────────────────────
→ (1) Downloading configuration...
config.json: 100%|██████████| 570/570 [00:00<00:00, 63.3kB/s]
→ (2) Downloading tokenizer...
tokenizer_config.json: 100%|██████████| 49.0/49.0 [00:00<00:00, 8.66kB/s]
vocab.txt: 100%|██████████| 213k/213k [00:00<00:00, 1.39MB/s]
tokenizer.json: 100%|██████████| 436k/436k [00:00<00:00, 10.1MB/s]
→ (3) Downloading model...
model.safetensors: 100%|██████████| 436M/436M [00:37<00:00, 11.6MB/s] 
✔ Successfully downloaded model "bert-base-cased"

── Downloading model "bert-large-uncased" ─────────────────────────────────────────
→ (1) Downloading configuration...
config.json: 100%|██████████| 571/571 [00:00<00:00, 268kB/s]
→ (2) Downloading tokenizer...
tokenizer_config.json: 100%|██████████| 48.0/48.0 [00:00<00:00, 12.0kB/s]
vocab.txt: 100%|██████████| 232k/232k [00:00<00:00, 1.50MB/s]
tokenizer.json: 100%|██████████| 466k/466k [00:00<00:00, 1.99MB/s]
→ (3) Downloading model...
model.safetensors: 100%|██████████| 1.34G/1.34G [01:36<00:00, 14.0MB/s]
✔ Successfully downloaded model "bert-large-uncased"

── Downloading model "bert-large-cased" ───────────────────────────────────────────
→ (1) Downloading configuration...
config.json: 100%|██████████| 762/762 [00:00<00:00, 125kB/s]
→ (2) Downloading tokenizer...
tokenizer_config.json: 100%|██████████| 49.0/49.0 [00:00<00:00, 12.3kB/s]
vocab.txt: 100%|██████████| 213k/213k [00:00<00:00, 1.41MB/s]
tokenizer.json: 100%|██████████| 436k/436k [00:00<00:00, 5.39MB/s]
→ (3) Downloading model...
model.safetensors: 100%|██████████| 1.34G/1.34G [01:35<00:00, 14.0MB/s]
✔ Successfully downloaded model "bert-large-cased"

── Downloading model "distilbert-base-uncased" ────────────────────────────────────
→ (1) Downloading configuration...
config.json: 100%|██████████| 483/483 [00:00<00:00, 161kB/s]
→ (2) Downloading tokenizer...
tokenizer_config.json: 100%|██████████| 48.0/48.0 [00:00<00:00, 9.46kB/s]
vocab.txt: 100%|██████████| 232k/232k [00:00<00:00, 16.5MB/s]
tokenizer.json: 100%|██████████| 466k/466k [00:00<00:00, 14.8MB/s]
→ (3) Downloading model...
model.safetensors: 100%|██████████| 268M/268M [00:19<00:00, 13.5MB/s] 
✔ Successfully downloaded model "distilbert-base-uncased"

── Downloading model "distilbert-base-cased" ──────────────────────────────────────
→ (1) Downloading configuration...
config.json: 100%|██████████| 465/465 [00:00<00:00, 233kB/s]
→ (2) Downloading tokenizer...
tokenizer_config.json: 100%|██████████| 49.0/49.0 [00:00<00:00, 9.80kB/s]
vocab.txt: 100%|██████████| 213k/213k [00:00<00:00, 1.39MB/s]
tokenizer.json: 100%|██████████| 436k/436k [00:00<00:00, 8.70MB/s]
→ (3) Downloading model...
model.safetensors: 100%|██████████| 263M/263M [00:24<00:00, 10.9MB/s] 
✔ Successfully downloaded model "distilbert-base-cased"

── Downloading model "albert-base-v1" ─────────────────────────────────────────────
→ (1) Downloading configuration...
config.json: 100%|██████████| 684/684 [00:00<00:00, 137kB/s]
→ (2) Downloading tokenizer...
tokenizer_config.json: 100%|██████████| 25.0/25.0 [00:00<00:00, 3.57kB/s]
spiece.model: 100%|██████████| 760k/760k [00:00<00:00, 4.93MB/s]
tokenizer.json: 100%|██████████| 1.31M/1.31M [00:00<00:00, 13.4MB/s]
→ (3) Downloading model...
model.safetensors: 100%|██████████| 47.4M/47.4M [00:03<00:00, 13.4MB/s]
✔ Successfully downloaded model "albert-base-v1"

── Downloading model "albert-base-v2" ─────────────────────────────────────────────
→ (1) Downloading configuration...
config.json: 100%|██████████| 684/684 [00:00<00:00, 137kB/s]
→ (2) Downloading tokenizer...
tokenizer_config.json: 100%|██████████| 25.0/25.0 [00:00<00:00, 4.17kB/s]
spiece.model: 100%|██████████| 760k/760k [00:00<00:00, 5.10MB/s]
tokenizer.json: 100%|██████████| 1.31M/1.31M [00:00<00:00, 6.93MB/s]
→ (3) Downloading model...
model.safetensors: 100%|██████████| 47.4M/47.4M [00:03<00:00, 13.8MB/s]
✔ Successfully downloaded model "albert-base-v2"

── Downloading model "roberta-base" ───────────────────────────────────────────────
→ (1) Downloading configuration...
config.json: 100%|██████████| 481/481 [00:00<00:00, 80.3kB/s]
→ (2) Downloading tokenizer...
tokenizer_config.json: 100%|██████████| 25.0/25.0 [00:00<00:00, 6.25kB/s]
vocab.json: 100%|██████████| 899k/899k [00:00<00:00, 2.72MB/s]
merges.txt: 100%|██████████| 456k/456k [00:00<00:00, 8.22MB/s]
tokenizer.json: 100%|██████████| 1.36M/1.36M [00:00<00:00, 8.56MB/s]
→ (3) Downloading model...
model.safetensors: 100%|██████████| 499M/499M [00:38<00:00, 12.9MB/s] 
✔ Successfully downloaded model "roberta-base"

── Downloading model "distilroberta-base" ─────────────────────────────────────────
→ (1) Downloading configuration...
config.json: 100%|██████████| 480/480 [00:00<00:00, 96.4kB/s]
→ (2) Downloading tokenizer...
tokenizer_config.json: 100%|██████████| 25.0/25.0 [00:00<00:00, 12.0kB/s]
vocab.json: 100%|██████████| 899k/899k [00:00<00:00, 6.59MB/s]
merges.txt: 100%|██████████| 456k/456k [00:00<00:00, 9.46MB/s]
tokenizer.json: 100%|██████████| 1.36M/1.36M [00:00<00:00, 11.5MB/s]
→ (3) Downloading model...
model.safetensors: 100%|██████████| 331M/331M [00:25<00:00, 13.0MB/s] 
✔ Successfully downloaded model "distilroberta-base"

── Downloading model "vinai/bertweet-base" ────────────────────────────────────────
→ (1) Downloading configuration...
config.json: 100%|██████████| 558/558 [00:00<00:00, 187kB/s]
→ (2) Downloading tokenizer...
vocab.txt: 100%|██████████| 843k/843k [00:00<00:00, 7.44MB/s]
bpe.codes: 100%|██████████| 1.08M/1.08M [00:00<00:00, 7.01MB/s]
tokenizer.json: 100%|██████████| 2.91M/2.91M [00:00<00:00, 9.10MB/s]
→ (3) Downloading model...
pytorch_model.bin: 100%|██████████| 543M/543M [00:48<00:00, 11.1MB/s] 
✔ Successfully downloaded model "vinai/bertweet-base"

── Downloading model "vinai/bertweet-large" ───────────────────────────────────────
→ (1) Downloading configuration...
config.json: 100%|██████████| 614/614 [00:00<00:00, 120kB/s]
→ (2) Downloading tokenizer...
vocab.json: 100%|██████████| 899k/899k [00:00<00:00, 5.90MB/s]
merges.txt: 100%|██████████| 456k/456k [00:00<00:00, 7.30MB/s]
tokenizer.json: 100%|██████████| 1.36M/1.36M [00:00<00:00, 8.31MB/s]
→ (3) Downloading model...
pytorch_model.bin: 100%|██████████| 1.42G/1.42G [02:29<00:00, 9.53MB/s]
✔ Successfully downloaded model "vinai/bertweet-large"

── Downloaded models: ──

                           size
albert-base-v1            45 MB
albert-base-v2            45 MB
bert-base-cased          416 MB
bert-base-uncased        420 MB
bert-large-cased        1277 MB
bert-large-uncased      1283 MB
distilbert-base-cased    251 MB
distilbert-base-uncased  256 MB
distilroberta-base       316 MB
roberta-base             476 MB
vinai/bertweet-base      517 MB
vinai/bertweet-large    1356 MB

✔ Downloaded models saved at C:/Users/Bruce/.cache/huggingface/hub (6.52 GB)

BERT_info(models)

                      model   size vocab  dims   mask
                     <fctr> <char> <int> <int> <char>
 1:       bert-base-uncased  420MB 30522   768 [MASK]
 2:         bert-base-cased  416MB 28996   768 [MASK]
 3:      bert-large-uncased 1283MB 30522  1024 [MASK]
 4:        bert-large-cased 1277MB 28996  1024 [MASK]
 5: distilbert-base-uncased  256MB 30522   768 [MASK]
 6:   distilbert-base-cased  251MB 28996   768 [MASK]
 7:          albert-base-v1   45MB 30000   128 [MASK]
 8:          albert-base-v2   45MB 30000   128 [MASK]
 9:            roberta-base  476MB 50265   768 <mask>
10:      distilroberta-base  316MB 50265   768 <mask>
11:     vinai/bertweet-base  517MB 64001   768 <mask>
12:    vinai/bertweet-large 1356MB 50265  1024 <mask>

(Tested 2024-05-16 on the developer’s computer: HP Probook 450 G10 Notebook PC)

While the FMAT is an innovative method for the computational intelligent analysis of psychology and society, you may also seek for an integrative toolbox for other text-analytic methods. Another R package I developed—PsychWordVec—is useful and user-friendly for word embedding analysis (e.g., the Word Embedding Association Test, WEAT). Please refer to its documentation and feel free to use it.