Elamur Research⌕
☰
home / papers / 2011 / Beyond Normalization: Pragmatics of Word Form in Text Messages

Beyond Normalization: Pragmatics of Word Form in Text Messages

Paper · 2011 · Tyler Baldwin and Joyce Chai

Summary

Normalization systems for text messages rewrite non-standard spellings such as "Yesssss" or "lookin" into dictionary forms. Baldwin and Chai argue that this step can erase meaning, because writers often choose an unusual spelling on purpose to add emotion or emphasis. To measure how common this is, they transcribed short conversations from screenshots on the "Damn You Autocorrect" website. They then extracted tokens missing from the CMU pronouncing dictionary and had Amazon Mechanical Turk workers judge each token in its dialogue context. About two fifths of the forms carried emotion absent from the standard spelling, mostly happiness. A similar share carried emphasis. Annotators also tended to see expressiveness, rather than saving keystrokes, as the writer's motive. The authors then framed a binary task: does a given non-standard form carry extra pragmatic content? An SVM using only character-level features clearly beat a majority baseline. Adding punctuation and position features gave no further gain. The work matters for text-to-speech and for any pipeline that cleans noisy text before interpreting it.

Key points

  • Data: 1190 conversations were transcribed. Out-of-vocabulary tokens were found with the CMU pronouncing dictionary, and real words such as names were removed by hand. This left 764 non-standard tokens.
  • Annotation: three MTurk workers judged each token on four questions:
    • the standard form,
    • any added emotion (Ekman's six basic emotions, plus none or other),
    • any other added information (friendliness/closeness, emphasis, other, none),
    • the writer's likely motive. Pairwise agreement was about 80%. Gold labels required at least two annotators to agree, and items without agreement were discarded separately for each question.
  • Findings: 40% of forms added emotion, with happiness dominant and fear never seen. 38% added emphasis and 20% signalled friendliness or closeness. Displaying extra information was the most commonly perceived motive.
  • Classifier: an SVM evaluated with ten-fold cross-validation. The majority baseline scored 59.5%. Character features alone reached 72.4% (p < 0.01), and adding punctuation (72.3%) or all features (72.4%) did not help.
  • Character features included:
    • letter repeated more than twice, and the maximum repeat count,
    • whether collapsing repeats yields an in-vocabulary word,
    • insertion-only edit distance as a cue for shortening,
    • whether the form is longer than its normalized version,
    • digits or symbols, concatenated words, and all-caps.

Interesting ideas

  • I1. Normalization should keep what it removes: the gap between surface spelling and standard form is a signal, not noise to throw away.
  • I2. A cheap elongation test works well: collapse repeated letters, check whether the result is in the lexicon, and record the longest repeat run.
  • I3. Annotators were asked why the writer deviated, separating expressive intent from economy and from plain error. This distinction turns spelling into evidence about attitude.
  • I4. Gold labels came from agreement by at least two of three judges, and disagreement was dropped per question rather than per item.
  • I5. The evaluation pairs a majority-class baseline with a feature ablation showing which feature groups add nothing. This keeps claims honest about where the signal actually comes from.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

  • I1 (tokens): when the en tokenizer maps "goooood" to the lemma "good", it could attach a form-mark attribute (elongated, all-caps, g-dropped) that later stages can read.
  • I2 (tokens): this is a transparent, rule-like morphology check that can resolve out-of-vocabulary forms without any learned model.
  • I3 (frames): prag could treat expressive spelling as an explainable cue for emphasis or positive stance, and use it when weighing speech acts or possible irony.
  • I4 (learn): silver data from witness parsers could keep the agreed fields of a sentence (for example heads) while discarding only the disputed ones (for example labels), instead of rejecting the whole sentence.
  • I5 (learn): each new feature group should be reported against a trivial baseline and an ablation, as a negative control for the gain it claims.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓Michigan State UniversityEast Lansing · United States7 works · 635 citations

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 9,085 worksArtificial Intelligence · 8,888 worksArts and Humanities · 9,296 worksLanguage and Linguistics · 9,206 worksHuman-Computer Interaction · 61 works

🧭 Topics 1Research subjects and languages the work is about.Sentiment & emotion · 609 works

Cite

Cite as Tyler Baldwin and Joyce Chai. 2011. Beyond Normalization: Pragmatics of Word Form in Text Messages. Proceedings of 5th International Joint Conference on Natural Language Processing.

The paper itself, not this page. BibTeX from the library's record of the work.

@inproceedings{baldwin2011beyond,
  title = {{Beyond Normalization: Pragmatics of Word Form in Text Messages}},
  author = {Tyler Baldwin and Joyce Chai},
  booktitle = {Proceedings of 5th International Joint Conference on Natural Language Processing},
  year = {2011},
  url = {https://aclanthology.org/I11-1169/}
}

References

18 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Apoorv Agarwal, Fadi Biadsy, and Kathleen Mckeown. 2009. Contextual phrase-level polarity analysis using lexical affect scoring and syntactic N-grams. In Proceedings of the 12th Conference of the European Chapter of the ACL (EACL 2009), pages 24– 32, Athens, Greece, March. Association for Computational Linguistics. find
  2. Cecilia Ovesdotter Alm, Dan Roth, and Richard Sproat. 2005. Emotions from text: machine learning for text-based emotion prediction. In Proceedings of the conference on Human Language Technology and Empirical Methods in Natural Language Processing, HLT ’05, pages 579–586, Stroudsburg, PA, USA. Association for Computational Linguistics.
  3. AiTi Aw, Min Zhang, Juan Xiao, and Jian Su. 2006. A phrase-based statistical model for sms text normalization. In Proceedings of the COLING/ACL on Main conference poster sessions, pages 33–40, Morristown, NJ, USA. Association for Computational Linguistics.
  4. Richard Beaufort, Sophie Roekhaut, Louise-Amélie Cougnon, and Cédrick Fairon. 2010. A hybrid rule/model-based finite-state framework for normalizing sms messages. In Proceedings of the 48th Annual Meeting of the Association for Computational Linguistics, pages 770–779, Uppsala, Sweden, July. Association for Computational Linguistics. find
  5. H. Binali, Chen Wu, and V. Potdar. 2010. Computational approaches for emotion detection in text. In Digital Ecosystems and Technologies (DEST), 2010 4th IEEE International Conference on, pages 172 – 177. find
  6. D.B. Bracewell. 2008. Semi-automatic creation of an emotion dictionary using wordnet and its evaluation. In Cybernetics and Intelligent Systems, 2008 IEEE Conference on, pages 1385 –1389. find
  7. Monojit Choudhury, Rahul Saraf, Vijit Jain, Animesh Mukherjee, Sudeshna Sarkar, and Anupam Basu. 2007. Investigation and modeling of the structure of texting language. Int. J. Doc. Anal. Recognit., 10(3):157–174. ↗
  8. Paul Cook and Suzanne Stevenson. 2009. An unsupervised model for text message normalization. In Proceedings of the Workshop on Computational Approaches to Linguistic Creativity, pages 71–78, Boulder, Colorado, June. Association for Computational Linguistics. find
  9. Paul Ekman. 1993. Facial expression and emotion. American Psychologist, 48:384–392. ↗
  10. E.C.-C. Kao, Chun-Chieh Liu, Ting-Hao Yang, Chang-Tai Hsieh, and Von-Wun Soo. 2009. Towards text-based emotion detection a survey and possible improvements. In Information Management and Engineering, 2009. ICIME ’09. International Conference on, pages 70 –74. find
  11. Catherine Kobus, François Yvon, and Géraldine Damnati. 2008. Normalizing SMS: are two metaphors better than one ? In Proceedings of the 22nd International Conference on Computational Linguistics (Coling 2008), pages 441–448, Manchester, UK, August. Coling 2008 Organizing Committee. find
  12. D.L. Pennell and Yang Liu. 2010. Normalization of text messages for text-to-speech. In Acoustics Speech and Signal Processing (ICASSP), 2010 IEEE International Conference on, pages 4842 –4845. find
  13. Changqin Quan and Fuji Ren. 2010. An exploration of features for recognizing word emotion. In Proceedings of the 23rd International Conference on Computational Linguistics (Coling 2010), pages 922– 930, Beijing, China, August. Coling 2010 Organizing Committee. find
  14. Richard Sproat, Alan W. Black, Stanley F. Chen, Shankar Kumar, Mari Ostendorf, and Christopher Richards. 2001. Normalization of non-standard words. Computer Speech and Language, pages 287–333. ↗
  15. Chung-Hsien Wu, Ze-Jing Chuang, and Yu-Chung Lin. 2006. Emotion recognition from text using semantic labels and separable mixture models. 5:165–183, June. ↗
  16. Changhua Yang, Kevin Hsin-Yih Lin, and Hsin-Hsi Chen. 2007. Emotion classification using web blog corpora. In Proceedings of the IEEE/WIC/ACM International Conference on Web Intelligence, WI ’07, pages 275–278, Washington, DC, USA. IEEE Computer Society. find
  17. Li Zhang, John A. Barnden, Robert J. Hendley, and Alan M. Wallington. 2006. Developments in affect detection in e-drama. In Proceedings of the Eleventh Conference of the European Chapter of the Association for Computational Linguistics: Posters &#38; Demonstrations, EACL ’06, pages 203–206, Stroudsburg, PA, USA. Association for Computational Linguistics. find
  18. Conghui Zhu, Jie Tang, Hang Li, Hwee Tou Ng, and Tiejun Zhao. 2007. A unified tagging approach to text normalization. In Proceedings of the 45th Annual Meeting of the Association of Computational Linguistics, pages 688–695, Prague, Czech Republic, June. Association for Computational Linguistics. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/baldwin-2011-beyond-normalization-pragmatics-word-form/