Normalization of non-standard words

By Richard Sproat, Alan W. Black, Stanley F. Chen, Shankar Kumar, Mari Ostendorf, Christopher D. Richards. 2001

397 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Natural Language Processing, Computational Linguistics & Semantics

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

In Computational Linguistics & Semantics

Citations in the corpus

Librispeech: An ASR corpus based on public domain audio books (2015, methodology): “Each book’s text is normalized by converting it into upper-c ase, removing the punctuation, and expanding common abbreviation s and non-standard words [8].”
A Survey on Neural Speech Synthesis (2021, methodology): “The text analysis module first processes the text, including text normalization [311], grapheme-to-phoneme conversion [24], word segmentation, etc, and then extracts the linguistic features, such as phonemes, duration and POS tags from different granularities.”
CVSS Corpus and Massively Multilingual Speech-to-Speech Translation (2022, neutral): “Nonstandard words (Sproat et al., 2001), such as numbers, currency expressions, dates, common abbreviations, acronyms, etc.”
Lexical Normalisation of Short Text Messages: Makn Sens a #twitter (2011, background): “Typos, ad hoc abbreviations, phonetic substitutions, ungrammatical structures and emoticons abound in short text messages, causing grief for text processing tools (Sproat et al., 2001; Ritter et al., 2010).”
Text normalization with convolutional neural networks (2018, neutral): “In numerous prior works mostly hand-constructed rules are used, and these rules are adjusted to specific domains of these systems (Sproat et al. 2001).”
What to do about bad language on the internet (2013, background): “Writing before social media was a signiﬁcant focus for NLP, Sproat et al. (2001) proposed to replace non-362 standard words with “the contextually appropriate word or sequence of words.””
Automated directory assistance system - from theory to practice (2007, methodology): “Since normalization rules are context dependent, it is usually conducted using a finite state transducer (FST) [11], with which the output symbols are determined based on the state and the input symbols.”
Neural Text Normalization with Subword Units (2019, background): “Non-standard words (NSWs) include expressions such as time or date (e.g., 4:58AM, 08-02, 8/2/2018), abbreviations (e.g., ft.) and letter sequences (e.g., IBM, DL) (Sproat et al., 2001).”
Analysis of named entity recognition and linking for tweets (2014, background): “Normalisation is a commonly proposed solution in cases where it is necessary to overcome or reduce linguistic noise (Sproat et al., 2001).”
Neural Models of Text Normalization for Speech Applications (2019, background): “Table 1 A taxonomy of non-standard words, from Sproat et al. (2001), Table 1, page 293.”
Unsupervised Learning for Expressive Speech Synthesis (2017, methodology): “There are also automatic approaches, like for instance by Sproat et al. (2001), where non-standard words are treated systematically on a language model basis, learning the pronunciation of tokens.”
Automatic Building of Synthetic Voices from Audio Books (2010, methodology): “Machine learning models such as Classification and Regression Trees (CART) are used to predict the class of NSW which is typically followed by rules to generate appropriate expansion of a NSW into a standard form [Sproat et al., 2001].”
The Kestrel TTS text normalization system (2014, background): “Again, following (Sproat et al. 2001) we call these non-standard words – NSWs .”
A Hybrid Text Normalization System Using Multi-Head Self-Attention For Mandarin (2019, methodology): “Currently, based on the traditional taxonomy approach for NSW[1], the Mandarin TN tasks are generally resolved by rule-based systems which use keywords and regular expressions to determine the SFW of ambiguous words[2, 3].”
RNN Approaches to Text Normalization: A Challenge (2016, background): “In the original written form there are two non-standard words (Sproat et al., 2001), namely the two measure expressions 6ft and 150lb .”
Unsupervised learning for text-to-speech synthesis (2013, background): “Finally, it should be noted that this thesis deliberately avoids broaching the issue of normalisation of non-standard words (Sproat et al., 2001), such as the normalisation of £5 to five pounds.”
The effects of part–of–speech tagging on text–to–speech synthesis for resource–scarce languages (2010, methodology): “The classification techniques employed in the normalisation and disambiguation processes range from simple regular expression rules to more elaborate context-sensitive rewrite rules, decision lists, decision trees and Naive Bayes classifiers [80, 83, 95].”