Latent Sequence Decompositions
Not on the map. Why?
Abstract (arXiv)
We present the Latent Sequence Decompositions (LSD) framework. LSD decomposes sequences with variable lengthed output units as a function of both the input sequence and the output sequence. We present a training algorithm which samples valid extensions and an approximate decoding algorithm. We experiment with the Wall Street Journal speech recognition task. Our LSD model achieves 12.9% WER compared to a character baseline of 14.8% WER. When combined with a convolutional network on the encoder, we achieve 9.6% WER.
Key facts. Cited by 6 papers in the map (data through 7 Oct 2026). First posted on arXiv 10 Oct 2016.
Builds on
No reference list yet.
Built on by
Citing papers per year: 2018: 1; 2019: 0; 2020: 0; 2021: 3; 2022: 0; 2023: 2.
6 papers.
- 2021Between words and characters: A Brief History of Open-Vocabulary Modeling and Tokenization in NLPcited by 130
- 2023End-to-End Speech Recognition: A Surveycited by 127
- 2021Acoustic Data-Driven Subword Modeling for End-to-End Speech Recognitioncited by 11
- 2018CompILE: Compositional Imitation Learning and Executioncited by 7
- 2021Do End-to-End Speech Recognition Models Care About Context?cited by 3
- 2023SelfSeg: A Self-supervised Sub-word Segmentation Method for Neural Machine Translationcited by 1
- 2023SelfSeg: A Self-supervised Sub-word Segmentation Method for Neural Machine Translation31 Jul 2023
- 2023End-to-End Speech Recognition: A Survey3 Mar 2023
- 2021Between words and characters: A Brief History of Open-Vocabulary Modeling and Tokenization in NLP20 Dec 2021
- 2021Acoustic Data-Driven Subword Modeling for End-to-End Speech Recognition19 Apr 2021
- 2021Do End-to-End Speech Recognition Models Care About Context?17 Feb 2021
- 2018CompILE: Compositional Imitation Learning and Execution4 Dec 2018