Danish Dependency Treebank

Annotation guide: Prepositions and adverbs

Matthias T. Kromann
Department of Computational Linguistics
Copenhagen Business School

Line Mikkelsen
Department of Linguistics
University of California, Santa Cruz

Prepositions

Prepositions are a closed word class consisting of the following words (the parentheses indicate the number of occurrences in the PAROLE corpus):

i (7196) over (491) bag (86) indtil (34) udover (7) ovenpå (1)
til (3757) mod (317) ad (68) foran (33) undtagen (6) henad (1)
på (3733) under (299) per (66) uanset (21) udenfor (6) iblandt (1)
af (3496) mellem (237) siden (58) overfor (20) nær (5) udfra (1)
med (3005) uden (217) før (56) bortset_fra (16)inklusive (5) opad (1)
for (2846) hos (166) inden (55) indenfor (15) angående (3) henover (1)
om (1445) gennem (151) trods (52) foruden (14) apropos (3) kontra (1)
fra (1196) omkring (134) igennem (43) forbi (12) vedrørende (2)  
ved (703) blandt (95) via (37) langs (11) førend (2)  
efter (554) ifølge (95) imod (37) imellem (11) nedover (1)  

Tests

Tests for determining whether an adverb and a following PP form a phrase, or if they modify the verb separately:

Edges

Edge type Edge label Description
complement nobj nominal object
vobj verbal object
pred predicative complement
adjunct mod modifier

Nominal complements Nominal complements to prepositions, including complements headed by "at", are tagged as nobj:

i haven         uden at falde

The analysis of "at"-complements as nobj is consistent with their analysis as dobj when occurring as complements to verbs. If the analysis of "at"+infinitive is changed, it should be changed across the board.

Verbal complements Finite clause complements without "at" are tagged as vobj:

uden de faldt        

Anternatively, finite clauses without "at" could be analyzed as nobj, where the verbal head has undergone syntactic type-shifting to a nominal (in analogy with the analysis of certain uses of adjectives as type-shifted nouns). This would simplify the valency frame of prepositions by excluding vobj. On the other hand there does not appear to any independent evidence for type-shifting with finite clauses, as there could be said to be in the case of adjectives. And there are distributional constrasts between bare finite clauses and finite clauses with "at". For instance, bare finite clauses cannot occur in subject position, whereas clauses with "at" can:

If bare finite clauses are (type-shifted into) nominals, this contrast has to be encoded by specific selection restrictions on the subject. If bare finite clauses are verbal, it follows from the generalization that verbs cannot be subjects.

Predicative complements Predicative complements are found in so-called absolutive constructions, headed by the prepositions "med" or "under". The preposition takes a nominal object (nobj) and a predicative complement (pred), which may be a preposition, adjective or noun. Note that a nominal predicative complement is preceded by "som" (as), which we analyze as the head of the predicative complement.

uden Susan på holdet taber vi         vi taber med ham som formand

Alternatively, the predicative complement could be analyzed as a dependent of the noun:

uden Susan på holdet taber vi

However, this would be (one of the) only context where nouns could take predicative complements. In ordinary predicative constructions a verbal head is required: There is a reading of "Uden Susan på holdet" where "på holdet" is a dependent of "Susan", but then it is a modifier, which helps establish the referent of Susan:

uden Susan på holdet taber vi

The analysis of "som" as the head of the predicative complement carries over to the use of "som" + nominal as (adverbial) modifiers:
Is this a different "som"? Which means something like "qua"? If so, do we need to distinguish the two syntactically?

Modifiers

Prepositions can take premodifiers (nouns, adjectives or adverbs). These are tagged as mod.

to minutter i syv         lige bag skolen

to minutter i syv

For several reasons, this analysis might not be appropriate for time expressions like "to minutter i syv". First, "to minutter" is not optional in this construction, as the modifiers in the other prepositional constructions are: Second, the external distribution of the whole phrase "to minutter i syv" is not that of a prepositional phrase, but closer to that of a noun: To accomodate these facts, one could proposen to analyze the initial numeral as the head of the construction. Since we analyze numerals as nominals, this would make the construction a nominal with respect to its external syntax.

to minutter i syv

The following contrast would still need to be accounted for:

Individual prepositions

FOR Occurs with full range of complements.
What type of obj is "vidt"?

"Vidt" is analysed as avobj in the corpus.

  • i stedet for

    i stedet for

  • Adverbial uses (tagged as RG): "For" can be used as a premodifier to adjectives, which takes an optional complement headed by "til" :
    • As an alternative one could analyze "for" as a preposition that takes an adjectival complement:

    De fik alt for lidt søvn

    One reason to prefer the modifier analysis is that "for" + adjective can also occur as an attributive modifier to a nominal, which is a role that adjectives, but not prepositions, can normally take:

    MED Occurs with full range of complements.

    When "med" occurs as a complement to "være", it could alternatively be analyzed as pred. However, "med" does not occur with all copula verbs, which speaks against the pred analysis:

  • As a particle "med" can cooccur with prepositional complements:

    På Occurs with full range of complements.

    The alternative is to analyze "ud" as a modifier to "på" in analogy with "oppe på".

  • De afleverede opgaven sidst på ugen

    De afleverede opgaven sidst på ugen

    Here we analyze "sidst" as a modifer to the preposition in analogy with the analysis of time expression like "to minutter i syv". As discussed here this might not be the correct analysis. In any case, the two constructions ("sidst på ugen" and "to minutter i syv") should be analyzed the same way.

  • Particle constructions

    I Occurs with full range of complements.

    Perhaps these are not special after all, and should instead be covered in a general section on the external syntax of prepositional phrases?

    This example presents several difficult, and partly interrelated, analytical choices:

    1. Is "rundt" a dependent of the verb or a dependent of the preposition? Note that both "De kørte rundt" and "De kørte i landet" are possible.

    2. Is the preposition (and it's dependents) a modifier to the verb or a complement?

    3. If it is a complement is it a regular pobj, or a `locative-directional complement' of the sort used in the Dutch treebank (see the discussion at the beginning of the chapter in verbs).

  • Fixed expressions

  • Particle constructions

  • Other constructions:

    TIL Occurs with full range of complements.

    Alternatively, "fra" could be analyzed as the head of the construction:

    fra før til nu

    The question is what type of complements "før" and "nu" are. They are tagged as adverbs (RG) in the corpus (in this use), but prepositions do not normally take adverbs as complements.

  • en gang til, tre meter til, to point fra eller til     [analyzed as a (coordinated) modifier]

    en gang til         to point fra eller til

  • med ryggen til, med kager til     [analyzed as absolutives]

    med ryggen til

  • Fixed expressions

  • Particle constructions

    AF Occurs with full range of complements.

    OVER Occurs with full range of complements.

    Alternatively, "over" could be analyzed as a modifier to "landet", but this miss the fact that "landet over" has the distribution of a prepositional phrase and not that of a noun:

  • {til/med/på/i/for} over tyve {kroner/timer}     ["over" as modifier to nobj]

    i over tyve timer

  • Han blev klar over problemet.     ["over" as pobj to adjective]

    han blev klar over problemet

  • fem minutter over     ["over" as modifier to nominal]

    fem minutter over

    Alternatively, "to minutter" could be analyzed as a (obligatory) modifier to "over". This would be more in line with the analysis of time expressions like "fem minutter over syv", though this analysis has some problems of its own, as discussed above.

  • format ud over det sædvanlige

    format ud over det sædvanlige

    Alternatively, one could analyze "ud over" (along with "ned over", "inden for", "uden for" etc.) as a complex preposition, which takes "det" as its nobj.

  • Fixed expressions

  • Particle uses

    OM Occurs with full range of complements.

    FRA Occurs with full range of complements.

    VED Occurs with full range of complements.

    EFTER Occurs with full range of complements.

    MOD Occurs with full range of complements.

    UNDER Occurs only with nominal complements!

    MELLEM Occurs, at least in principle, with all types of complements

    UDEN Occurs with full range of complements.

    HOS Occurs with nominal complements only!

    GENNEM Seems to occur only with nominal complements

    OMKRING Seems to occur only with nominal complements (but there is a change from "om" -> "omkring" which widens the complementation of "omkring")

    BAG Occurs only with nominal complements

    RUNDT Can behave as a preposition, when "rundt" occurs with a locational meaning:

    De kørte landet rundt

    as opposed to a time-indicating meaning:

    De fester året rundt

    The reasons for choosing a different analysis for the locational and time-indicating meanings are:

    OTHERS: blandt (75) ad (68) pr. (63) siden (53) før (53) på_grund_af (52) ifølge (49) inden (46) igennem (41) trods (38) via (37) imod (34) foran (33) Foran (2) indtil (22) overfor (20)

    Adverbials

    Expressions with verb + adverbial + PP

    In the set of sample analyses all occurrences, but one (in ex 71), are analyzed in terms of pobj's as follows (cf. example 23):

    Hun satte sig frem i stolen

    There are two questions: The first question is largely theoretical. As I see it, the alternatives is to analyze the adverb as a particle (part), where it forms a close connection with the verb, and as a modifier (mod) where it doesn't.

    The answer to the second question seems to be negative: this is not a uniform construction, though we might decide to gloss over the differences in the treebank analyses, for reasons of simplicity and manageability. One can distinguish at least the following types of construction:

    The marked paragraph above was written before the labels avobj and lobj came into this manual-world. These two labels solve some of the problems discussed above. All of the examples in the paragraph above would now be analysed as lobj, except "gå ud over": this would be analysed as avobj + pobj. Below follow some general remarks:

    See under verbs for principles demarkating avobj/lobj/part/mod and pobj.

    Time-indicating objects in relative indications of time

    Adverbial use

    Time indications that are relative to a fixed point in time all follow a certain pattern in construction. They are headed by an adverb, an adjectiv (used as an adverb) or a preposition (used as an adverb), e.g. før, siden, tilbage, tidligere. Some can take an object, in which case they are "converted into" a preposition or a subordinating conjunction (CS), if the object is a sentence (e.g. før/før brylluppet/før festen begyndte). They all allow a preceeding modification by a time specification, e.g. "to timer før". This time-specification is analysed as a time-indicating object (tobj).

    The adverbs/prepositions that allow this construction, are:

    Examples:

    Hun kom hjem to dage inden festen begyndte

    Det ligger flere uger tilbage i tiden

    Det skete to timer tidligere end de havde regnet med

    Det er to år siden de kom hjem

    Notice the special analysis of the fixed expression "for ... siden":

    Huset blev bygget for tre år siden

    Nominal use

    The analyses above apply only to adverbial uses. The same adverbs combined with relative indications of time allow a nominal use also. Take a look at these contrastive examples:

    Here 1) is an example of adverbial use, while 2) shows a nominal use, where the phrase is the subject of the sentence. These different kinds of use mirror a different underlying meaning, which we think should be reflected by the analyses: When playing the role of an adverb, the adverb is the head of the phrase, taking an optional time-indicating object; when playing the role of a noun, the noun is the head of the phrase, thus:

    Han kom hjem året før Året før var det bedste år i mit liv

    The nominal use demands that the noun is definite; the adverb must be "inden", "forinden", "før" or "efter".

    Special constructions with "ikke"

    Så

    Like in the discussion of "hvor" in "hun ved, hvor svært det kan være" (see the discussion under verbs , relatives) we have been troubled by examples with "så AN ..., at ..." ("so AN ... that ..."). Here we have chosen to analyse "så" as a modifier to the AN, and "at" as a nominal object to "så", like this:

    Han var så lykkelig at han græd

    For analysis of "så længe + [sentence]", see the paragraph "Relatives" under Verbs.

    Obligatory modifiers

    Sometimes prepositional phrases and adverbials occur in a sentence in such a way that a) one tends to analyse it as a modifier since different types of phrases can occur here, but at the same time b) it is not optional in the way modifiers normally are. We have decided to label these "obligatory modifiers" (edge: obl). Here are some examples:

    Both "end" and "som" are a little bit like gapping constructions: "Det er bedre end i skolen" could be analysed as a short form of "Det er bedre end det er i skolen". In that case the analysis should look like this:

    Det er bedre end i skolen

    In spite of the plausibility of this analysis we have decided to let the analysis of "end" and "som" be parallel to the other constructions; but the eliptical character should be kept in mind.

    http://www.id.cbs.dk/~stine/manual/preps.html last updated by Stine Kern Lynge at 2003-11-12 10:11