[matrix] Re: [rmrs] tense and mood info in rmrs

[Anette Frank]

Hi,
thanks, Dorothee and Lars, for your further comments and examples.

Here some more observations on where I see problems in mapping
shallow, purely morphologically or constructionally steered TAM features
to a full-fledged and fine-grained tense and aspect marking:
- TELIC or STATIVE: If viewed as representations for lexical aspect
(Aktionsarten), we cannot map to these categories in languages where
these are not morphologically marked, if we only have a non-lexicon-driven
shallow parser.
- As Dorothee already mentioned, the aspectual contribution of perfect
in German is difficult to determine from morphological cues. In general,
past perfect vs. imperfect tense in German do not correspond to clear
aspectual distinctions.
- It might be possible to represent progressive for English 
appropriately - in
German we don't mark progressive. Though I am not sure whether
prog+ in English can be reliably mapped to PROTRACTED+ STATIVE-
in English without further knowledge about lexical aspect classes - 
again from
the viewpoint of shallow parsing. (Examples I have in mind are things like:
He was being enthousiastic - if this is correct English).

So I agree with your conclusion, that one way to proceed could be to 
preserve
the TAM features for the encoding of morphological info for shallower
systems that cannot make these distinctions.
It also would seem a viable (intermediary) solution as long as it is not 
yet clear
how the some of the TAM markings used in some of the HPSG grammars
are to be translated to RMRS encoding (see Melanies email).

My main concern was to retain TENSE information, including aspect (in
the Reichenbachian sense) whereever possible - and the German and English
(French and Spanish alike) systems, at least, make this possible to a large
extent, even for a shallow parser.
If we went for a TAM encoding of TENSE, we could first of all ignore all 
the
complications involved in phenomena like sequence of tense, etc. that
complicate a simple mapping from morpho-syntactic marking to purely 
semantic
TENSE features, and I think we could retain much information that a shallow
parser can deliver, that will be useful for futher temporal processing in a
hybrid architecture.

If we could agree on this proposal, what should the TAM-encoding look like
and where should it find it's place in the RMRS?
Should it correspond to what the grammars (G/E/?others) currently deliver?
i.e.
present ->TENSE=present
simple past -> TENSE=past
present perfect -> TENSE=present, PERF=+
past perfect -> TENSE=past, PERF=+
simple future -> TENSE=future
future perfect -> TENSE=future, PERF=+

I have had a look into the matrix files, but haven't found a 
specification about
the proposed features and their combination in tam. Ideally, if there 
were a
TAM encoding available in RMRS, it should correspond to the feature system
that's used by the Matrix grammars, I would think.

Best, Anette


Dorothee Beermann wrote:

> This from Ann's last mail:
>
>
>
>> It may not be possible with the current feature set, but could we 
>> establish
>> some examples that demonstrate the problem?
>
>
> Here two examples, one that shows that 'progressives' in more 
> fine-grained aspectual system
> than English and German can be covered with the present 
> rmrs-specifications
> (repeated below), while the other example shows the opposite, namely 
> that the distinctions for
> perfective expression made in the Kwa languages of West-Africa could 
> not be covered.
>
> But before we turn to the examples, we feel like repeating that our 
> impression concerning the
> present discussion is that for most practical purposes in the work 
> with Indo-European
> languages, the present DTD specifications are sufficient ( but see 
> more about Anette's concern below).
>
> Now to the examples which we think are important as they give a taste 
> of what will
> be needed for the future of multi-lingual grammar engineering. So, 
> back to AKAN:
>
> Akan distinguishes:
>
> (i)  progressive vs. continuative
> and
> (ii) perfective vs. completive
>
> (i) progressive is a specification of events while continuative 
> applies to protracted states.
> (ii) perfective is a completed event that dates back, using reference 
> time as the anchor
> while completive is not specified for tempus.
>
> Given the features below from the rmrs DTD we can express (i) as:
>
> PROTRACTED + STATIVE -    versus   PROTRACTED+ STATIVE +
>
> (ii) would come out as:
>
> TELIC +  TENSE past(?)    versus  TELIC +  TENSE ??
>
> (ii) is the interesting case, since in order to express the difference 
> between
> (a) perfective and (b) completive we have to say that (a) is 
> referentially anchored
> with respect to tempus but (b) is not. With the system that we have, 
> where tempus is
> tense and as such so far best suited to cover morphological 
> distinctions like those made in English we potentially fall short of 
> what we would like to express cross-linguistically.
>
> rmrs-DTD
> =============
> tense (past|present|future|non-past|u) #IMPLIED
> telic (plus|minus|u) #IMPLIED
> protracted (plus|minus|u) #IMPLIED
> stative (plus|minus|u) #IMPLIED
> incept (plus|minus|u) #IMPLIED
> imr (plus|minus|u) #IMPLIED
> boundedness (plus|minus|u) #IMPLIED
> refdistinct (plus|minus|u) #IMPLIED >
>
> =================
> Anette writes:
> It would therefore seem to me a viable solution to record 
> morphological tense features in the
> RMRS, alongside the semantic tense and aspect classes, in order to 
> record basic tense
> and mood distinctions from shallow parsers, where available. The 
> alternative would be to throw
> away this information altogether, which doesn't seem reasonable to me.
>
> We agree that throwing away information seems like a bad idea. 
> Retaining it, as it comes delivered through the shallow parser - would 
> that amount to introducing a new feature under
> TAM, such as MORPH-TAM?
>
> It seems that even inside German, the category 'perfect' is not 
> semantically unambiguous considering the following examples:
>
> (iii)  Getrieben vom Schicksal  wird er ziellos durch die Strassen 
> wandern.
> (iv)  So gepackt wirst du mit dem Koffer nicht weit kommen.
>
> For (iii) and (iv)  the connection to the semantic tense/aspect 
> features is less than straightforward
> and perhaps harder to do then for the run of the mill things like:
>
> (v)   Er hatte den Koffer gepackt
>
> where the perfect-marked verb is part of a periphrastic tense-aspect 
> marking system.
>
> best
> Lars and Dorothee
>