Antonyms in NLP Applications

Antonyms and negation are essential for many NLP applications as described below:

• Negation detection
  • Negation detection has generated special interest in extra-propositional aspects of meaning in text in practical NLP applications. Especially in systems processing medical and clinical text, such as outpatient notes, patient records, etc.
  • NLP applications have been developed to extract clinical information from medical records. The most common types of information extracted are diagnoses or findings. Identifying the negation status of a finding is as important as identifying the finding itself [2011 Goryachev]. For example, a finding occurring in a negated context may indicate the absence of some medical condition. Search tools looking for documents containing a particular finding may return irrelevant results if they do not take the negation into account (misleading).
  • Extra-propositional aspects of meaning in text exploring different aspects of meaning such as factivity (Saurı and Pustejovsky, 2009), uncertainty/hedging (Farkas et al., 2010), committed belief (Prabhakaran et al., 2010), and modalities (Prabhakaran et al., 2012a). Among these, negation detection has generated special interest because of demonstrated needs for negation detection capability in practical applications such as information retrieval (Averbuch et al., 2004), information extraction (Meystre et al., 2008), sentiment analysis (Wiegand et al., 2010; Councill et al., 2010), and relation detection (Chowdhury and Lavelli, 2013). Accurately detecting negations is especially important in systems processing medical and clinical text.
  • In fact, most phrases indicating negation are stop words in information retrieval systems and are not even used for indexing. Is negative indexing helpful and needed?
  • Concept mapping: Example: [Mild hyperinflation without focal pneumonia]: “without” is important from this patient’s clinical record. It indicates the absence of focal pneumonia in the patient. Not capturing this extra-propositional aspect of negation concerning focal pneumonia will lead to wrong and harmful inferences in downstream processing, e.g. by a clinical decision support system.
  • In NLP, cTake, CLAMP or Metamap, negation is detected to denote whether a given concept is absent or present [2017 Manimaran, 2001 Chapman]. But, no further mapping.

• Antonyms
  • Antonym detection has applications in tasks of understanding language, such as paraphrase detection and generation, or contradiction detecting.
    • detecting and generating paraphrases:
      => [The dementors caught Sirius Black / Black could not escape the dementors]
    • detecting contradictions:
      => [Kyoto has a predominantly wet climate / It is mostly dry in Kyoto]

  • Some antonym pairs include negative terms, which can be used for negation detection.

  • Paraphrase or paraphrasing in computational linguistics is the natural language processing task of detecting and generating paraphrases. Applications of paraphrasing are varied including: question answering, machine translation, sentiment analysis (SA) and information retrieval (Roth and Schulte im Walde, 2014; Mohammad et al., 2013) and textual inference.
    • Sentiment Analysis: in Sentiment Analysis the correct discrimination of antonyms (e.g. good from bad) is extremely important to identify the positive or negative polarity of a text. [2015 Enrico Santus]. For example, words of the same and opposing polarity need to be distinguished [2014 Roth]
    • Textual entailment: need to identify hypernymy because of directional inference requirements. [2014 Roth]
    • information retrieval, [2004 Averbuch], [2013 Mohammad]
    • question answering:
    • text summarization:
    • plagiarism detection:
    • Machine translation: [2013 Mohammad]
    • Dialogue systems: [2013 Mohammad]
    • Information extraction: [2008 Meystre]
    • Relation detection: [2010 Weigand, 2010 Councill]
    • Clinical question answering [2006 Lee]
    • Clinical decision support [2009 Demner-Fushman]
    • Medical information extraction [2010 Uzner]
    • Patient history tracking [2012 Tymoshenko]