status of words; HPSG and CG

[Andreas Kathol]

Dear Dean,

> I have a question about the status of words and the different feature
> complexes in HPSG and Construction Grammar.
>
> As I understand it, HPSG assumes strict lexicalism: word structure and
> phrase structure are governed by independent principles (a quote taken right
> from the Stanford web page).  CG

I would like to inject a word of caution here re. the term
"Construction Grammar". In my experience, this term does not denote a
uniform (or even internally consistent) set of assumptions. The brand
that Goldberg represents (which makes explicit reference to radial
categories, metaphors, on-line default-override operations) is in many
ways distinct from the version of Construction Grammar that has been
developed by Fillmore and Kay. The latter is quite closer in spirit
(and often in actual analyses) to HPSG, especially those recent
proposals that have taken a construction-based view of
grammar. Therefore, Construction Grammar, understood as uniting people
like Goldberg and Fillmore/Kay under a common label, has always struck
me as a sociological artifact. In terms of actual linguistic practice,
Fillmore/Kay/Sag form a natural class, often to the exclusion of
Goldberg.

> makes a different assumption (Goldberg,
> 1995, p.7):
>
> "In Construction Grammar, no strict division is assumed between the lexicon
> and syntax.  Lexical constructions and syntactic constructions differ in
> internal complexity, and also in the extent to which phonological form is
> specified, but both lexical and syntactic constructions are essentially the
> same type of declaratively represented data structure: both pair form with
> meaning.  It is not the case, however, that in rejecting a strict division,
> Construction Grammar denies the existence of any distinctly morphological or
> syntactic constraints (or constructions).  Rather, it is claimed that there
> are basic commonalities between the two types of constructions, and
> moreover, that there are cases, such as verb-particle combinations, that
> blur the boundary."
>
> Following Goldberg, I wonder if it is reasonable to include, within a
> feature complex, combinatorial features of word structure that are parallel
> to combinatorial features of phrase structure.

In my opinion, the Goldberg quote makes no commitment as to whether
syntactic combination should be driven by the same mechanisms as
morphological combination. In fact, one could (and many people do)
assume that the "lack of strict division" between phrasal and lexical
information pertains to the fact that both essentially consist of
pairings of sound and meaning and some morphsyntactic categorization,
while the "distinctly morphological [vs.]  syntactic constraints"
pertain precisely to the ways that larger units are formed from
smaller ones in syntax vs. morphology.

For example, consider noun
> incorporation structures in Cree (an Algonquian language spoken by about
> 100,000 people in northern Canada):
>
> (1) pahpaw-ahow-iiw    mistatimwa
> brush-IS-IS    horse	(IS = inflectional suffix – oversimplified here)
> ‘He brushes the horse(s).’
>
> (2)  pahpaw–astimw–ii–w
> brush-horse-IS-IS
> ‘He brushes the horse(s).’
>
> In (1), the NP that follows the verb could be represented by a syntactic
> feature (e.g., subcategorization, valence), as in (3) (a rather simple
> feature complex):
>
> (3)
> PHON <pahpaw>
> POS <Verb>
> SYN <NP>
>
> In (2), the polysynthetic noun incorporation might be represented by a
> parallel morphological feature, as in (4):
>
> (4)
> PHON <pahpaw>
> POS <Verb>
> MORPH <N>     (there are reasons to think it is a noun rather than an NP)

Why not simply describe _pahpaw_ in terms of a *lexeme* that licenses
a nominal dependent? Then one word-formation schema may interpret this
as saying that the nominal has to be a phrasal dependents (probably an
optional one at that) and you get (1), while another word-formation
schema deriving noun-incorporated structures may interpret the
"proto-valence" information of the lexeme in terms of a nominal stem
which forms part of a larger morphological unit, as in (2).

> If the feature complex in (4) was feasible, we might also propose something
> similar for inflectional suffixes.  Thus, the English phrase
> ‘walked to the
> store’ might be represented by (5):
>
> (5)
> PHON <walk>
> POS <Verb>
> MORPH <PAST>
> SYN <PP>
>
> The sign <PAST> would include regular suffixes.  Internal change and
> suppletion would require something else.

I don't quite see what would be gained by stating the affixes in terms
of a separate feature. In any case, inflectional morphology strikes me
as rather different from noun-incorporation.

> Given these possible feature complexes, I am interested in comments and
> references that address issues such as:
> - in CG and HPSG, are the assumptions about the status of words
> reconcilable?
> - what arguments exist against including features of morphological
> combination within signs/feature structure?  How do these arguments relate
> to data and proposals such as these? (e.g. what do we do about irregular
> morphology?)

It seems to me that the issue you're alluding to is one that has been
debated in morphology more generally under the headings
"item-and-arrangement" vs. "item-and-process". The kind of arguments
that one may marshall against a pure syntacticization of morphology
are precisely those that advocates of the "item-and-process" view have
developed. Suppletion and irregular morphology have been adduced
precisely to argue in favor of the item-and-process view.

> - are there previous HPSG analyses of morphological structure that provide
> insight here?

You may want to look at Krieger and Nerbonne's (1993?) treatment of
_-bar_ affixation in terms of a morphological combinatorial system
that parallels that of syntax. The disadvantages of such an approach
have been discussed extensively by Susanne Riehemann. More recently,
Stefan Mueller (talk at Trondheim HPSG conference) has proposed a
treatment of German separable prefixes in which valence information is
either "cashed out" syntactically (prefix as a separate syntactic
unit) or morphologically (prefix as part of a lexical word). His work
may come closer to what you're interested in because is looks like it
could provide a model of how to deal with the Cree cases.

Best,

  --Andreas