Text-Token Predictor Feature

A Text-Token Predictor Feature is a predictor feature function that is based on information about a text-token.

• Context:
  • It can range from being a Language-Dependent Predictor Feature to being a Language-Independent Predictor Feature.
  • It can (typically) be a Categorical Predictor Feature.
  • It can (often) be used as:
    • an Text Tagging Predictor Feature, such as a POS Feature.
    • a Text Segmentation Predictor Feature, such as a Concept Mention Identification Feature.
• Example(s):
  • a Low-level Text Token Predictor Feature.
    • a Text Token hasCapitalLetter Feature, such as [math]\displaystyle{ f }[/math](hasCapital("Markov”)) ⇒ 1
    • a Text Token Dictionary Match Feature, such as [math]\displaystyle{ f }[/math](equals("Markov”,"Jordan”)) ⇒ 0
    • a Text Token Character Pattern Feature, such as [math]\displaystyle{ f }[/math](charPattern("Machine-223") ⇒ "Aaaaaaa-000" (Collins, 2002).
    • a Character n-Gram Feature, such as [math]\displaystyle{ f }[/math](“rko”, “Markov”) ⇒ true.
  • a High-level Text Token Predictor Feature.
    • a Text Token Part-of-Speech Role Feature, such as [math]\displaystyle{ f }[/math](POS(NNS, “tokens”) ⇒ 1
• Counter-Example(s):
  • Text Window-based Predictor Feature.
  • Token Sequence Predictor Feature.
• See: NER Predictor Feature.

---

---