Corpora: SUM: metrical scanner

From: Constantina Stamou (
Date: Wed Mar 20 2002 - 16:25:32 MET

  • Next message: LDC Office: "Corpora: New Release from the LDC"

    Dear list members,

    this is what I received on my query about metrical scanners for English verse:

    Thank you very much for your responses.


    From: Jason Eisner <>

    Not aware of anything, but such a system could be built.

    The first step is to find the stressed syllables when the poem is read
    as prose. This might be approximated by looking up the dictionary
    pronunciations of the words. In general, however, the stress pattern
    sometimes depends on part of speech ("progress," "content," etc.) and
    even on context.

    Text-to-speech systems have to solve this problem, and you may be
    able to use such a system to convert your poem to a phonological
    representation that accurately marks syllables and their stresses.

    Now you have to get from the stress pattern to a tag indicating the
    meter type. Hand-written heuristics may suffice for this, though
    sloppy meter is perhaps harder to recognize than strict meter. Another
    option would be to annotate some examples with tags and train a
    statistical model on them. For example, an HMM whose state records
    the last few observations (stress, no-stress, line break) as well as
    the hidden current tag.

    -cheers, jason eisner

    From: Susan Hockey <>

    You can find some discussion of computers and metrical
    scansion in my book Electronic Texts in the Humanities,
    OUP 2000. The success of this technique depends to
    a large extent on the natural language and the specific metre.

    Susan Hockey

    This archive was generated by hypermail 2b29 : Wed Mar 20 2002 - 16:46:19 MET