The structure of early/earlier/later/further on

[BrettRey]

Could this be something like ago where you have a head preposition that requires modification?

• A while ago, we found another example.
• Later on, we find another example.
• Further on/down/over, we find another example.
• A little way on/down/over, we find another example.
• *A little way, we find another example.
• *On/down/over, we find another example.
• A while later/ago/back/after, we found another example.
• Later/*Ago/*Back/*After, we found another example.
• Later/Further on, we find another example.
• Later/*Further, we find another example. [Further is ungrammatical if it means "as we proceeded"; of course grammatical if "furthermore".]
• Earlier on, we found another example.
• Earlier, we found another example.
• Later still, we found another example.
• *Still, we found another example. [Ungrammatical unless "still" has a different meaning.]
• Further on/down/over/up/away/afield/north, we find another example.
• Down/Up/Away/?North/*On/*Over/*Afield, we find another example.
• A little way on/down/over/up/away/afield/north, we find another example.
• long/#far ago
• #long/far back

[nschneid]

For the later still example, it seems clear that later is the head and still is a modifier.

For the other cases, I suspect there's a paper to be written. :)

Three days on/in, the experiment was showing no signs of success. Like a little way on etc. It's an extent + direction construction. Neither the extent nor the direction can be omitted.

Earlier on, etc.: the preposition can be omitted, but not "earlier". Arguably this means "earlier" is the head and "on" a complement in the AdvP.

CGEL p. 708 notes: "The complement of from often contains on or onwards (from that time on/onwards)"

Early/late + PP: see also UniversalDependencies/UD_English-EWT#390

[nschneid]

With ago, I'm not convinced by the "mandatory modifier" analysis in CGEL. Everywhere else, mandatory dependents are called complements. Clearly ago/hence/back are unlike most prepositions; it seems simple enough to say they have an exceptional (postpositional) order w.r.t. their complement.

"Ago", unlike the others, cannot be premodified by right/just (a moment just before/*ago).

OTOH, I guess we have to consider how this interacts with other constructions. Consider extraction/stranding:

• They will get married in a year.
  ?How long will they get married in?
  In how long will they get married?
  the amount of time they will get married in
• They got married after a year.
  ?How long did they get married after?
  How long after did they get married?
  the amount of time they got married after
• She was born a year before (they got married).
  *How long was she born before (they got married)?
  How long before (they got married) was she born?
  *the amount of time she was born before (they got married)
• They got married a year ago.
  *How long did they get married ago?
  How long ago did they get married?
  *the amount of time they got married ago

In the above cases "X ago" patterns more like "X before" than "in/after X" (where X is the duration).

If we were to go with the CGEL analysis that the duration before "ago" is not a complement, then I would rather treat its function as simply "dependent" as opposed to "modifier".

[BrettRey]

To be clear, the mandatory modifier analysis is SIEG2's. CGEL and SIEG1 have a pre-head complement analysis.

[nschneid]

Oops OK

[BrettRey]

Long looks like a modifier and not an object in every case except before long

1. long after ; *after long
2. long ago ; *ago long
3. long past that time ; *past long
4. long into the night ; *into long
5. long in advance ; *in long
6. long before ; before long

[BrettRey]

Why, though, shouldn't shortly work with ago as a modifier?

[nschneid]

I am wondering if long is exceptional.

Shortly or significantly can precede before, but not ago. Same with much.

Note also expressions where long occurs where an NP like "a long time" would normally be expected: for long, before long (p. 640), and it won't take long. p. 569:

[nschneid]

CGEL p. 574 also looks relevant