Improved Part-of-Speech Tagging for Online Conversational Text with Word Clusters

By Olutobi Owoputi, Brendan T. O'Connor, Chris Dyer, Kevin Gimpel, Nathan Schneider, Noah A. Smith. 2013

818 citations recorded by the source corpus.

How researchers see this paper

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

In Natural Language Processing

Citations in the corpus

Gaussian Error Linear Units (GELUs) (2016, methodology): “To meet this challenge we compare the nonlinearities on POS-annotated tweets (Gimpel et al., 2011; Owoputi et al., 2013) which contain 25 tags.”
A Baseline for Detecting Misclassified and Out-of-Distribution Examples in Neural Networks (2016, methodology): “…nonlinearity, a weight initialization according to (Hendrycks & Gimpel, 2016c), pretrained word vectors trained on a corpus of 56 million tweets (Owoputi et al., 2013), and a hidden layer size of 256, all while training on 1000 tweets for 30 epochs with Adam and early stopping with 327…”
Demographic Dialectal Variation in Social Media: A Case Study of African-American English (2016, methodology): “To search for the constructions, we tagged the corpora using the ARK Twitter POS tagger (Gimpel et al., 2011; Owoputi et al., 2013), 9 which Jørgensen et al. (2015) show has similar accuracy rates on both AAE and non-AAE tweets, unlike other POS tag-gers.”
Adversarial Removal of Demographic Attributes from Text Data (2018, methodology): “Preprocessing We tokenize each tweet using twokenize , a twitter speciﬁc tokenizer (O’Connor et al., 2010; Owoputi et al., 2013), and discard duplicate tweets and tweets with less than three to-kens.”
Snips Voice Platform: an embedded Spoken Language Understanding system for private-by-design voice interfaces (2018, methodology): “More specifically, we use Brown clusters [9] released by the authors of [37], as well as word clusters built (6)BILOU is a standard acronym referring to the possible positions of a symbol in a sequence: Beginning, Inside, Last, Outside, Unit.”
Tagging Performance Correlates with Author Age (2015, methodology): “Our POS taggers are trained using TreeTagger with default parameters, and CRF++ with default parameters and standard POS features (Owoputi et al., 2013; Hovy et al., 2015b).”
Learning a POS tagger for AAVE-like language (2016, background): “Word representations learned from representative unlabeled data, such as word clusters or embeddings, have been proven useful for increasing the accuracy of NLP tools for low-resource languages and domains (Owoputi et al., 2013; Aldarmaki and Diab, 2015; Gouws and Søgaard, 2015).”
Tailoring Continuous Word Representations for Dependency Parsing (2014, background): “In addition to work mentioned above, relevant work that uses discrete representations exists for POS tagging (Ritter et al., 2011; Owoputi et al., 2013), named entity recognition (Ratinov and Roth, 2009), supersense tagging (Grave et al., 2013), grammar induction (Spitkovsky et al., 2011),…”
FuzzE: Fuzzy Fairness Evaluation of Offensive Language Classifiers on African-American English (2020, methodology): “We use a Twitter-speciﬁc part-of-speech tagger, Twokenizer (Owoputi et al. 2013) to annotate each tweet.”
Modelling input texts: from Tree Kernels to Deep Learning (2015, neutral): “pos-tagger [51,104] to obtain the part-of-speech tags.”
Twitter Universal Dependency Parsing for African-American and Mainstream American English (2018, background): “(This is inspired by Scherrer (2011)’s rule transforms be-tween Standard and Swiss German.)”
Investigating Sports Commentator Bias within a Large Corpus of American Football Broadcasts (2019, methodology): “5 As part-of-speech tags predicted by spaCy are unreliable on our transcript text, we tag FOOTBALL using the ARK TweetNLP POS tagger (Owoputi et al., 2013), which is more robust to noisy and fragmented text, including TV”
Stylistic Variation in Social Media Part-of-Speech Tagging (2018, methodology): “We choose Twitter because language styles in this platform are particularly diverse (Eisenstein et al., 2010), and because moderately large labeled datasets are available (Gimpel et al., 2011; Owoputi et al., 2013).”
Learning Approximate Inference Networks for Structured Prediction (2018, methodology): “To obtain training data for training the tag language model, we run the Twitter POS tagger from Owoputi et al. (2013) on a dataset of 303K randomly-sampled English tweets.”
Predicting crime using Twitter and kernel density estimation (2014, methodology): “We addressed these problems using the Twitter-specific tokenizer and part-of-speech tagger developed by Owoputi et al. [22].”