New benchmark PhoPunct evaluates Transformer architectures for Vietnamese punctuation restoration

Researchers have developed PhoPunct, a large-scale benchmark for evaluating Vietnamese punctuation restoration models. The benchmark, comprising 22.1 million labeled tokens from News and Novels corpora, was used to test 17 different Transformer-based architectures. Findings indicate that adding Bi-LSTM and CRF layers does not consistently improve performance under a fixed training budget, with plain Transformer classifiers often matching or exceeding their Bi-LSTM-CRF counterparts. PhoPunct achieved the best results on the News corpus, while XLM-R with a CRF performed best on the Novels corpus. AI

IMPACT Provides a standardized evaluation for punctuation restoration models, potentially guiding future research in NLP for Vietnamese.

RANK_REASON The cluster describes a new academic paper introducing a benchmark and evaluating model architectures. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• arXiv
• bidirectional long short-term memory network and attention mechanism with semantic mining for e-commerce web page recommendation
• conditional random field
• Hugging Face
• PhoBERT-large-Bi-LSTM-CRF
• PhoPunct
• transformer
• Vietnamese
• XLM-RoBERTa

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →