Bilstm Armenian Participle Punct is a word-level bidirectional LSTM tagger designed for punctuation restoration in Eastern Armenian participle clauses. The model addresses the specific task of labeling punctuation within these clauses, using a four-class token labeling scheme to identify whether a token should be followed by a comma, a specific punctuation mark (BUTH_AFTER), have a comma removed, or remain unchanged.
This tool employs frozen Armenian GloVe embeddings with 300 dimensions and operates with a single-layer BiLSTM architecture, featuring a hidden size of 128. It is implemented as a custom PyTorch model rather than using transformer-based architectures. The model is trained and evaluated on both noisy web text and clean textbook data, with benchmark macro-F1 scores provided for these datasets. Tokenization is performed at the word level using a supplied vocabulary file.
json). The model and its resources are distributed under the MIT license, allowing for open-source use and modification.
Bilstm Armenian Participle Punct is suitable for tasks in token classification and punctuation restoration for Eastern Armenian, particularly in research or applications involving low-resource syntax sequence labeling.
Bilstm Armenian Participle Punct sits in PulseGate's Other AI category. It focuses on restoring punctuation in Armenian participle clauses using an open-source sequence labeling model. Bilstm Armenian Participle Punct is an open-source project aimed at NLP researchers and Armenian language technologists. The project is open source (Open Source). It runs on the web, the command line, and API, and it can be self-hosted.
It is developed by Albert Hakobyan, and the product first shipped in 2026. The project is developed in the open on GitHub with 2 commits in the last 90 days. Among its 5 catalogued features are punctuation restoration, token classification, and armenian language support.
Latest indexed changes and source events
AlbertHakobyan/bilstm-armenian-participle-punct verified by the PulseGate indexer
Other apps tracked under the same category.
