Over tapas on 17 April 2018, the discussion of arguments v. adjuncts turned up an interesting avenue to explore: How do the degrees of freedom in ERG lexical entries relate to the various tests for the argument/adjunct distinction. This page is meant as a place to collect (a) those degrees of freedom/dimensions of variation, (b) work towards understanding how they do and don't cross-classify (surely some of the cells in that multidimensional matrix are empty) and (c) how they relate to the various tests in the literature.
ERG Dimensions of variation
- Syntactically selected or not (aka "argument" v. "adjunct", in ERG terms)
- Semantic role: none (e.g. in raising constructions), semantic argument, semantic functor
- Subdimension for semantic functor: scopal or non-scopal
- Control relation or not
- Obligatory or not
Persistent tests in the literature
- Iterability: But maybe not with manner?
- Obligatoriness: Syntactically required
- Dialogue test: Semantically required
- Ontological obligatoriness: every event described by this verb requires a certain role
Specificity: Appears with only a few verbs, then it's probably an argument of those verbs. (Converse of universality or flexibility) Ex: selected PPs wait for somebody
Do so test: Dan ate sushi with Stephan's wife and I did so with somebody else's wife. #And I did so pizza. Not just about do so not being able to combine with an NP. *Kim relied on Chris on Tuesday and Sandy did so on Pat on Wednesday.
Less stable/more debunked tests
- Extraction only from arguments, not adjuncts
Rough notes from 24 April WG meeting
==== Illustrating ERG dimensions ====
Kim ate sushi in Oslo sushi: selected, assigned a role in Oslo: not selected, takes verb as an argument
Kim stumbled into the room: into the room: selected, takes verb as an argument
Kim jumped into the water: selected, but takes verb as an argument jump(kim) into(jump,water)
Kim slipped into the room. Kim slipped (different sense only).
Kim put the books in the box put(Kim,books,in the box) in(book,the box)
- Iterability -- assigned by different theories: HPSG, LFG, FGD (FGD uses it to decide whether to put it in the valence lexicon)
Misalignment between dialogue test & ontological obligatoriness --- time & place mostly ontologically required, but won't pass the dialogue the test.
Arrive -- where you arrive is obligatory, but not from where:
A: Charles arrived yesterday. B: Where did he arrive? A: #I don't know. A: Charles arrived yesterday. B: Where did he arrive from? A: I don't know. A: Charles arrived. B: When did he arrive? A: I don't know.
The marchers came from all corners of the city and numbered over 100,000 when they arrived at city hall.
She put the books in the box in the basement. -- only ''in the box'' fills the ARG3 of ''put''
They arrived in the train station in Oslo. In Oslo, they arrived in the train station. ?In the train station, they arrived in Oslo. He arrived in Europe yesterday in France in a small town called Pons. He lived in France for a long time in a small town called Pons.
Prague folks have a hard time picking which one is the argument and which is the adjunct.
Notes from ERG Walk-Through
More fine-grained bar-levels than ‘classic’ HPSG: MODIFIED feature (with internal structure) allows adjuncts to leave a mark on their head. Used for grammaticality contrasts in *We arrived the day vs. We arrived the first day. Also used to control spurious ambiguity with pre- and post-head modification. Maybe also used to block iteration, e.g. with multiple cardinal adjectives.