SRILM - an extensible language modeling toolkit

By Andreas Stolcke. 2002

5,178 citations recorded by the source corpus.

SRILM is a collection of C++ libraries, executable programs, and helper scripts designed to allow both production of and experimentation with statistical language models for speech recognition and other applications. SRILM is freely available for noncommercial purposes. The toolkit supports creation and evaluation of a variety of language model types based on N-gram statistics, as well as several related tasks, such as statistical tagging and manipulation of N-best lists and word lattices. This paper summarizes the functionality of the toolkit and discusses its design and implementation, highlighting ease of rapid prototyping, reusability, and combinability of tools.

Subfields: Artificial Intelligence, Natural Language Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Natural Language Processing

Citations in the corpus

The LAMBADA dataset: Word prediction requiring a broad discourse context (2016, methodology): “We implemented a simple RNN (Elman, 1990), a Long Short-Term Memory network (LSTM) (Hochreiter and Schmidhuber, 1997), a traditional statistical N-Gram language model (Stolcke, 2002) with and without cache, and a Memory Network (Sukhbaatar et al.”
Creating Training Corpora for NLG Micro-Planners (2017, methodology): “Three types of bi-gram models are extracted from category graphs using the SRILM toolkit (Stolcke, 2002): one model (S-Model) for bigrams occurring in sibling triples (triples with a shared subject); one model (C-Model) for bigrams occurring in chained triples (the object of one triple is the subjec”
Librispeech: An ASR corpus based on public domain audio books (2015, methodology): “Then the SRILM toolkit [9] is used to train a Witten-Bell [10] smoothed bigram language model on the tex t of that book.”
Attention-over-Attention Neural Networks for Reading Comprehension (2016, methodology): “All language model features are trained on the training proportion of each dataset, with 8-gram wordbased setting and Kneser-Ney smoothing (Kneser and Ney, 1995) trained by SRILM toolkit (Stolcke, 2002).”
Targeted Syntactic Evaluation of Language Models (2018, methodology): “N-gram model: We trained a 5-gram model on the same 90M word corpus using the SRILM toolkit (Stolcke, 2002) which backs off to smaller n-grams using Kneser-Ney smoothing.”
A fast and simple algorithm for training neural probabilistic language models (2012, methodology): “As a baseline for comparison, we also trained several n - gram models (with modiﬁed Kneser-Ney smoothing) using the SRILM toolkit (Stolcke, 2002), obtaining results similar to those reported by Zweig & Burges (2011).”
Quasar: Datasets for Question Answering by Search and Reading (2017, methodology): “We train three n-gram baselines using the SRILM toolkit (Stolcke et al., 2002) for n = 3 , 4 , 5 on the entire corpus of all Stack Overﬂow posts.”
Minimum Risk Training for Neural Machine Translation (2015, methodology): “M OSES uses the parallel corpus to train a phrase-based translation model and the target part to train a 4-gram language model using the SRILM toolkit (Stolcke, 2002).”
Achieving Human Parity in Conversational Speech Recognition (2016, methodology): “We use an N-gram language model trained and pruned with the SRILM toolkit [56].”
ESPnet: End-to-End Speech Processing Toolkit (2018, methodology): “1Language modeling is often performed by external language model toolkits, for example SRILM [6] 2https://github.”
Findings of the 2017 Conference on Machine Translation (WMT17) (2017, background): “, 2013) and a statistical language model (Stolcke, 2002).”
The microsoft 2016 conversational speech recognition system (2016, methodology): “We use an N-gram language model trained and pruned with the SRILM toolkit [32].”
The CAPIO 2017 Conversational Speech Recognition System (2017, methodology): “The 4-gram language model (LM) was trained with the open-source library of SRILM [31] on a combination of publicly available data, including Fisher English Training Part 1 and 2 (LDC2004T19, LDC2005T19), Switchboard-1 Release 2 (LDC97S62), CallHome American English Speech (LDC97T14), Switchboard Cel”
Improved training of end-to-end attention models for speech recognition (2018, methodology): “For LibriSpeech, we also train Kneser-Ney smoothed n - gram count based language models [53] on the same BPE vocabulary set using SRILM toolkit [54].”
Semi-Supervised Learning for Neural Machine Translation (2016, methodology): “We use the SRILM toolkit (Stolcke, 2002) to train 4-gram language models.”
TED-LIUM: an Automatic Speech Recognition dedicated corpus (2012, methodology): “First, we initiated the process by decoding all the available audio data using the default acoustic models provided in the CMU Sphinx 3 package and a 4-gram language model trained with the SRILM toolkit (Stolcke, 2002) on all the text contained in the closed captions.”
Compositional Morphology for Word Representations and Language Modelling (2014, methodology): “This includes a baseline 4-gram MKN language model, trained with SRILM (Stolcke, 2002) and queried efﬁciently using KenLM (Heaﬁeld, 2011).”
EU-BRIDGE MT: Combined Machine Translation (2014, methodology): “Language models are trained with the SRILM toolkit (Stolcke, 2002) and use modiﬁed Kneser-Ney smoothing.”