Fingerspelling - the sequential signing of letters that compose proper names, technical terms, and out-of-vocabulary words - is an integral component of any sign language. Despite Russian Sign Language (RSL) being used by more than 150,000 native signers, no public fingerspelling dataset for RSL exists, in contrast to American, British, German, and several other sign languages. We introduce Znaki the first open RSL fingerspelling dataset: 37,252 video recordings (35.4 hours) of 1,593 unique sequences - proper names sampled from 16 domain categories - recorded by 68 native RSL signers (37 Deaf, 29 hard-of-hearing, and 2 certified interpreters) through a crowdsourcing pipeline with a two-stage signer qualification procedure and expert validation. The dataset covers all 33 Cyrillic handshapes, and its letter-frequency distribution matches the Russian National Corpus. As reference points for future research, we establish two baselines adapted from American Sign Language fingerspelling recognition: a Squeezeformer-based encoder-decoder, reaching a character-level accuracy (1−CER) of 0.932 and a sequence accuracy of 0.596, and Fingerspelling-PoseNet, reaching 0.886 and 0.545, respectively.
Figures & tables
Figure 1: The 33 signs of the RSL alphabet, each corresponding to a letter of the Russian alphabet. Arrows indicate the direction of movement for 8 dynamic signs.
Dataset
Language
Samples
Signers
Hours
Type
RWTH Dreuw et al. (2006)
DGS
1,400
20
—
isolated
NSL23 Sunuwar et al. (2023)
NSL
630
14
—
isolated
Bukva Kvanchiani et al. (2024)
RSL
3,757
155
—
isolated
NSTU Grif and Kondratenko (2021)
RSL
14,978
4
—
isolated (private)
ChicagoFSWild Shi et al. (2019b)
ASL
7,304
168
—
continuous
ChicagoFSWild+ Shi et al. (2019a)
ASL
55,232
260
—
continuous
Table 1: Publicly available fingerspelling datasets. Znaki is the first continuous fingerspelling dataset for RSL.
Figure 2: Sample frames from the Znaki dataset.
Category
Count
Moscow Metro
231
Social terms
121
Banking terms
107
Countries
99
Russian cities
99
Authors
98
Table 2: The number of unique samples in each category of the Znaki dataset.
Figure 3: Task for recording the proper name «Виктор Гюго» (Victor Hugo).
Figure 4: Illustration of the annotation scheme for fingerspelling sequence boundaries (start and end).
Figure 5: Age and gender distribution in the Znaki dataset.
Figure 6: Comparison of letter frequencies in the Znaki dataset and the Russian National Corpus.
Figure 7: Znaki dataset statistics: a) video length distribution; b) distance distribution, where the distance to the camera is approximately estimated in meters by computing the length between the signer’s left and right shoulders using MediaPipe Lugaresi et al. (2019) .
Figure 8: Distribution of the total number of letters across the dataset splits.
Model
1−CER↑
Acc ↑
FS-PoseNet
0.886
0.545
Transformer-Based
0.932
0.596
Table 3: Baseline results on the proposed RSL fingerspelling dataset.