Skip to content

Latest commit

 

History

History
67 lines (48 loc) · 3.17 KB

annotations_eng.md

File metadata and controls

67 lines (48 loc) · 3.17 KB
Raw

For texts written in contemporary Swedish Sparv can generate the following types of annotations:

  • Part of speech tagging:

    • pos: part of speech
    • msd: morphosyntactic tag

    Tool: Hunpos
    Model: in-house model trained on SUC 3.0
    Tag set: MSD tags

  • SALDO-based analysis:

    • baseform: citation form
    • lemgram: lemgram, identifies the inflectional table (using SALDO tags)
    • sense: identifies a sense in SALDO and its probability
    • (saldo: identifies a sense in SALDO - will be removed soon)
    • sentiment: sentiment score using SenSALDO

  • Compound analysis (also based on SALDO):

    • complemgram: compound lemgram
    • compwf: compound word form
    • (prefix: initial part of a compound - will be removed soon)
    • (suffix: final part of a compound - will be removed soon)

  • Dependency analysis:

    • ref: the position of the word in the sentence
    • dephead: dependency head, the ref of the word which the current word modifies or is dependent of
    • deprel: dependency relation, the relation of the current word to its dependency head

    Tool: MaltParser
    Model: swemalt, trained on Swedish Treebank
    Tag set: Mamba-Dep

  • Named entity recognition:

    • ne.ex: named entity (name expression, numerical expression or time expression)
    • ne.type: named entity type
    • ne.subtype: named entity sub type

    Tool: hfst-SweNER
    References: HFST-SweNER – A New NER Resource for Swedish, Reducing the effect of name explosion

  • Readability metrics:

    • text.lix: the Swedish readability metric LIX, läsbarhetsindex
    • text.ovix: the Swedish readability metric OVIX, ordvariationsindex
    • text.nk: the Swedish readability metric nominalkvot

  • Lexical classes:

    • blingbring: lexical class from the Blingbring resource (on word level)
    • swefn: frames from swedish FrameNet (on word level)
    • text.blingbring: lexical class from the Blingbring resource (on document level)
    • text.swefn: frames from swedish FrameNet (on document level)

Older Swedish texts or texts written in other languages can often be annotated with a sub set of the above annotation types.

The msd annotation for non-Swedish languages is based on different tag sets, depending on which language is annotated and what annotation tool is being used. The attribute contains information about the part of speech and in many cases morphosyntactic information. The pos annotation contains only part-of-speech information and uses the universal POS tag set.