Skip to content

Latest commit

 

History

History
128 lines (93 loc) · 4.96 KB

guidelines.org

File metadata and controls

128 lines (93 loc) · 4.96 KB

Annotation Guidelines

Aspectual class

An event is characterised by a verb and its complements. Prototypically, the direct object of the verb represents the entity that is affected or changed by the verb’s action; the nature of this entity is important to the event’s aspectual class. For instance, the number of the direct object can determine the aspectual class of the event:

  • Ich aß einen Apfel. (dynamic, bounded, durative, change-of-state)
  • Ich aß viele Äpfel. (dynamic, unbounded)

The grammatical subjects of unaccusative verbs are semantic patients, affected by the expressed action, and must also be taken into consideration when determining the aspectual class of a verb.

  • consider resultative adjuncts
  • ignore aspectual coercion via adjuncts, except as far as these provide cues to the fundamental aspectual class

Valid

Verbs should be tagged as invalid if they cannot reasonably be assigned an aspectual class. Because the pre-processing of the annotation corpus is done automatically, the corpus may contain verbs which are not appropriate to the primary task. Examples include:

  • Mis-tagged verbs (e.g., auxiliaries which have been falsely classified as a main verb.
  • Sentence fragments, where the complements to a verb needed for the assessment of lexical aspect are not shown.

Stative

Stative verbs describe a state of being, rather than an action (process or event). Stative verbs express that a property obtains at a particular point or interval in time, and imply that this property holds over a (usually unspecified) duration. Unlike dynamic verbs, stative verbs cannot denote temporally bounded activities, and they do not directly express changes of state.

Stative verbs are frequently incompatible with the imperative mood.

  • Geh nach Hause!
  • ? Weiß die Antwort!

Stative verbs cannot be the complement of the verb to force:

  • Ich habe sie gezwungen, nach Hause zu gehen.
  • ? Ich habe sie gezwungen, die Antwort zu wissen.

Bounded

Bounded events have a natural conclusion, a point in time at which the expressed action is finished and cannot continue any longer. The use of the past tense implies a sense of culmination:

  • Ich habe einen Apfel gegessen.
  • Ich habe eine Sonata gespielt.
  • Ich lese ein Buch.

Unbounded events cannot be finished, although they may stop:

  • Ich habe Äpfel gegessen.
  • Ich habe Klavier gespielt.
  • Ich habe gelesen.

Adverbials of time (für eine halbe Stunde/in einer halben Stunde)

Punctual vs. Durative

  • iterative interpretation when combined with durative adverbials
    • Ich habe (für zehn Minuten/zehn Minuten lang) geschlafen.
    • Ich habe (für zehn Minuten/zehn Minuten lang) geniesst.
    • Ich habe (den ganzen Nachmittag) mein Zimmer geputzt.
    • Ich habe (den ganzen Nachmittag) meine Schlüssel gefunden.

Change of State

  • lasting change effected in the patient of the verb
    • Ich habe das Buch gelesen. (?)
    • Ich habe das Buch vorgelesen. (no change of state)
    • Ich habe den Studenten das Buch vorgelesen. (change of state)

Appendix: Construction of the verb corpus

We begin by collecting 60 verbs to annotate; 20 verbs are drawn at random from each of three groups: high frequency verbs (the first 65 verbs, with counts of over 10^5 in sdewac); medium frequency verbs (the next 602 verbs, with counts of over 10^4 in sdewac); and low frequency verbs (some 2100 more verbs, with counts between 10^3 and 10^4 in sdewac).

Polysemy of these verbs can be estimated by taking the number of senses listed in GermaNet for a particular verb.

[INSERT FIGURE]

We can see that the estimated polysemy is higher for the high frequency verbs, while the medium and low frequency verbs don’t seem to differ from each other very much. The high frequency group still contains one verb with only a single synset (entstehen), so this verb should be examined in detail to see what an unambiguous high frequency verb does. The distribution of fraction of instances co-occurring with aspectual indicators also resembles the distribution of the whole corpus. All this suggests that the sample is at least representative of the corpus.

We then collect 50 clauses for each of the 60 verbs (for a total of 3000 clauses), which contain the verb. We work with only clauses, so that there are examples in the corpus such as:

["die", "die", "Kommission", "an", "das", "Parlament", "gemacht", "hat", ".”].

In choosing these clauses, we select only from those where number of words per clause (sentence?) is less than (or equal to?) 50, and where there is no negation; we do not filter out non-finite verb forms.