jurafsky&martin_3rdEd_17 (1).pdf

Questionssuch as a rule extracting any noun phrase

Info icon This preview shows pages 413–414. Sign up to view the full content.

possible by the relatively stylized syntax of Jeopardy! questions—such as a rule extracting any noun phrase with determiner “this” as in the Conan Doyle example, and rules extracting pronouns like she , he , hers , him , as in the poet example. The lexical answer type (shown in blue above) is a word or words which tell lexical answer type us something about the semantic type of the answer. Because of the wide variety of questions in Jeopardy!, Jeopardy! uses a far larger set of answer types than the sets for standard factoid algorithms like the one shown in Fig. 27.4 . Even a large set of named entity tags is insufficient to define a set of answer types. The DeepQA team investigated a set of 20,000 questions and found that a named entity tagger with over 100 named entity types covered less than half the types in these questions. Thus DeepQA extracts a wide variety of words to be answer types; roughly 5,000 lexical answer types occurred in the 20,000 questions they investigated, often with multiple answer types in each question. These lexical answer types are again extracted by rules: the default rule is to choose the syntactic headword of the focus. Other rules improve this default choice. For example additional lexical answer types can be words in the question that are coreferent with or have a particular syntactic relation with the focus, such as head- words of appositives or predicative nominatives of the focus. In some cases even the Jeopardy! category can act as a lexical answer type, if it refers to a type of entity that is compatible with the other lexical answer types. Thus in the first case above, he , poet , and clerk are all lexical answer types. In addition to using the rules directly as a classifier, they can instead be used as features in a logisitic regression classifier that can return a probability as well as a lexical answer type. Note that answer types function quite differently in DeepQA than the purely IR- based factoid question answerers. In the algorithm described in Section 27.1 , we determine the answer type, and then use a strict filtering algorithm only considering text strings that have exactly that type. In DeepQA, by contrast, we extract lots of answers, unconstrained by answer type, and a set of answer types, and then in the later ‘candidate answer scoring’ phase, we simply score how well each answer fits the answer types as one of many sources of evidence. Finally the question is classified by type (definition question, multiple-choice, puzzle, fill-in-the-blank). This is generally done by writing pattern-matching regular expressions over words or parse trees. In the second candidate answer generation stage, we combine the processed question with external documents and other knowledge sources to suggest many candidate answers. These candidate answers can either be extracted from text docu- ments or from structured knowledge bases.
Image of page 413

Info icon This preview has intentionally blurred sections. Sign up to view the full version.

Image of page 414
This is the end of the preview. Sign up to access the rest of the document.

{[ snackBarMessage ]}

What students are saying

  • Left Quote Icon

    As a current student on this bumpy collegiate pathway, I stumbled upon Course Hero, where I can find study resources for nearly all my courses, get online help from tutors 24/7, and even share my old projects, papers, and lecture notes with other students.

    Student Picture

    Kiran Temple University Fox School of Business ‘17, Course Hero Intern

  • Left Quote Icon

    I cannot even describe how much Course Hero helped me this summer. It’s truly become something I can always rely on and help me. In the end, I was not only able to survive summer classes, but I was able to thrive thanks to Course Hero.

    Student Picture

    Dana University of Pennsylvania ‘17, Course Hero Intern

  • Left Quote Icon

    The ability to access any university’s resources through Course Hero proved invaluable in my case. I was behind on Tulane coursework and actually used UCLA’s materials to help me move forward and get everything together on time.

    Student Picture

    Jill Tulane University ‘16, Course Hero Intern