-
Notifications
You must be signed in to change notification settings - Fork 4
LADLushootseedSemantics
Lushootseed is a Central Salishan language spoken in the Puget Sound region of what is now Washington State, USA. One hypothesis about Salishan languages is that all verb roots are intransitive (Intransitivity Hypothesis; Davis and Matthewson, 2009), the corollary is that all transitive verbs are derived. This phenomenon brings up questions about how to perspicaciously model the semantic reflexion of such transitivization processes: should we provide structures which build up transitive predicates hand-in-hand with the morphological rules which derive them from intransitive bases? Or should we "swap out" one predicate for another in the application of transitivizing lexical rules? RMRS (Copestake n.d.,Copestake 2009) may provide a way to model the compositional structure of transitivizing lexical rules in a typed-feature structure based grammar. The purpose of this session is to present some data which motivate the use of RMRS and to discuss the application of RMRS principles in this context.
Quoting Beck (2009, 1--2):
"""
-
What are transitive verbs in most languages are derived from a large set of monovalent patient-oriented (Hess 1995) radicals whose syntactic subject expresses the semantic PATIENT or ENDPOINT of an event rather than the AGENT. Consider (1):
1. a. ʔuɬič̓ čəd. ʔu-ɬič̓ čəd PFV-be.cut 1SG.SUB `I got cut with a knife.' b. ʔuɬič̓id čəd tə sqʷiqʷali ʔu-ɬič̓i-t čəd tə sqʷiqʷali PFV-be.cut-ICS 1SG.SUB INDEF hay ‘I started to cut hay (with a blade)’
(Bates, Hess & Hilbert 1994: 146)
-
√ɬic̓ ‘be cut (with a knife)’—in spite of expressing an event high on the scale of semantic transitivity (Hopper & Thompson 1980)—can take only a single syntactic argument, a subject expressing the PATIENT of the event
-
to express an AGENT, it is necessary to apply a valency-increasing suffix such as the internal causative -t
"""
So, the Lushootseed lexicon has many such pairs of alternants where an underived verb has a single argument which seems to correspond most closely with what Dowty called proto-patient, and a derived transitive where both an argument in correspondence with proto-agent appears as well as one corresponding to proto-patient. This leads to questions about representations in MRS. One choice is to list the two verbs separately in a lexicon which is based on stems after derviational suffixes have applied:
ɬič̓=: intransitive-lex-item &
[ STEM < "ɬič̓">,
SYNSEM.LKEYS.KEYREL.PRED "_ɬič̓_v_be.cut_rel" ].
ɬič̓id =: transitive-lex-item &
[ STEM < "ɬič̓id">,
SYNSEM.LKEYS.KEYREL.PRED "_ɬič̓id_v_be.cut_rel" ].
where
intransitive-lex-item := basic-one-arg-no-hcons & basic-icons-lex-item &
[ ARG-ST < [ LOCAL.CONT.HOOK [ INDEX ref-ind & #ind,
ICONS-KEY.IARG1 #clause ] ] >,
SYNSEM [ LKEYS.KEYREL.ARG1 #ind,
LOCAL.CONT.HOOK.CLAUSE-KEY #clause ] ].
transitive-lex-item := basic-two-arg-no-hcons & basic-icons-lex-item &
[ ARG-ST < [ LOCAL.CONT.HOOK [ INDEX ref-ind & #ind1,
ICONS-KEY.IARG1 #clause ] ],
[ LOCAL.CONT.HOOK [ INDEX ref-ind & #ind2,
ICONS-KEY.IARG1 #clause ] ] >,
SYNSEM [ LKEYS.KEYREL [ ARG1 #ind1,
ARG2 #ind2 ],
LOCAL.CONT.HOOK.CLAUSE-KEY #clause ] ].
But this misses the generalization that the event-type denoted by _ɬič̓_v_be.cut_rel and _ɬič̓id_v_be.cut_rel are linguistically the same, we have two predicate strings which refer to the same event type but with different argument structures, ie, we've packed a syntactic dependency into a semantic predicate name.
Another option would be to treat the transitivizer in a way similar to the ERG's treatment of periphrastic causatives. In this scenario, the transitivizing lex rules add a _cause_x_rel. Traditionally, this _cause_x_rel has its own characteristic variable, and then takes two further args, one corresponding to the CAUSER (perhaps something like a proto-agent), the other will be the characteristic (event) variable of the root. Under this sort of analysis, a verb like ɬič̓id would have semantics:
RELS:
-
[ _ɬič̓_v_be.cut_rel,
- ARG0 e0 ARG1 x0 ],
[ _cause_x_rel,
- ARG0 e1 ARG1 e0 ARG2 x1 ]
We want this to mean something like x1 cut x0 with a knife.
One thing that stands out to me about this analysis is the "extra" e variable on the cause rel. That is, I don't yet see the motivation for this. I think one of the characteristics of the AGENT prototype is acting as a causer of an event. For example, the English verb "fell" means to cut down, but someone might argue that this is a morphological causative variant of "fall". The ERG demo gives _fell_v_rel(e,x0,x1) for fell in "Sandy felled the tree". If we are happy to analyze English transitives as denoting a single event, I'm not convinced that the presence of a derivational affix in Lushootseed is motivation enough to warrant a proliferation of event variables in the Lushootseed grammar.
A third idea for representing these structures comes from using Parsons-style decomposition of the arity such that:
- RELS which introduce event variables are always monadic, they predicate a string-label for the event and nothing more
- RELS for AGENT,PATIENT,GOAL,etc are bivalent, their signatures being something like this: REL-NAME(e,x).
Under this sort of system, (1a) looks like this:
-
[ _ɬič̓_v_rel,
- ARG0 e0 ],
[ _patient_rel,
- ARG0 e0, ARG1 x0 [ PNG 1sg ] ]
And (1b) looks almost exactly the same, but with the x0 being 3sg (some hay), and with one further REL added:
- [ _agent_rel
- ARG0 e0, ARG1 x1 [ PNG 1sg ] ]
In fact, our DELPH-IN formalism already has a precedent for treating argument relations as predicates a la Parsons, this is called R(obust)MRS. In RMRS, thematic role names aren't used such as _agent_rel, but instead the predicate names which are used are the familiar ARGN types, a la Dowty's ordered-argument/proto-role (Dowty 1989, 1990). Here I hasten to comment that in interpreting these ordered argument relations, the lexical class of the verb has to be taken into account, Lushootseed also has unergative monovalent roots (Hess' agent-oriented radicals) in addition to the unaccusative ones like ɬič̓. There are few more examples down below. EDIT: the more I think about this, the less sure I am of the previous comment---perhaps independent of verb class, ARG1,2,3 should be proto-agent, proto-patient, proto-goal/location, etc. In this case, there will be verb which can only select for an ARG2, for example.
Copestake motivates RMRS (Copestake n.d., 2009) as an MRS variant which is amenable to encoding the output of shallow processing (such as POS tagging) in a notion which is compatible with information from a deep grammar. The idea is that if you can write default rules which map POS tags onto semantic types, but one of the principle things you're missing in that scenario is the arity of the predicate for a given item. Parsons-style reification of argument relations as predicate names allows bare predicate names to be monads, predicated of an event variable and nothing more. RMRS then allows for syntactic dependencies to further predicate syntactic argument relations on that event, just in case you have a tool for finding them. Essentially, RMRS allows for underspecification of predicate-arity in this way. Although the motivation is different here, for Lushootseed, an underspecification of predicate arity allows that intransitive roots can their derived transitive variants share a predicate type.
Beyond the decomposition of predicate arity, yet connected to it, one thing which is different between the "standard" MRS and RMRS is that labels are required to be unique for each relation. Instead of using labels for conjunction at a particular node of the scope tree, RMRS adds explicit conjunction constraints to the representation. Each label, then, is used as the first argument to the ARGN_REL predicates. Keeping with the idea that argNs are to be interpreted with respect to verb class (so arg1 on unaccusatives refers to a proto-patient) and continuing the example above using ɬič̓id, it seems like we'd have something like the following for "ɬič̓id čəd tə sqʷiqʷali" (I cut the hay (with a knife).)
a0:_ɬič̓_v_rel(e1)
a2:_arg1_rel(a0, x3 [PNG 3sg])
a4:_arg2_rel(a0, x5 [PNG 1sg])
One question which pops out to me here, is why should I use the label a0, as the thing passed into the first argument of the ARGN rels rather than the actual event variable 'e1' in the example directly above? It would seem that using the e variable would allow us to bring back in the explicit coordination using labels, as it's the fact that each label has to stand in for its predicate's argument as a reference point which forces the requirement that labels be unique.
Beyond the descriptions of the RASP system in the Copestake papers and references therein, I would like to know if there are typed-feature structure grammars which have implemented RMRS. I've heard talk of an ERG branch which was using RMRS. Is this just an apocryphal tale? Are there versions of the LKB or PET which know what to do with in-g constraints when calculating the scope-machinery? For me, seeing some of this stuff in use would be very interesting and useful.
- What collection of previous DELPH-IN work should I download and look at for previous RMRS work?
- What DELPH-IN tools are compatible with RMRS format?
- What subfeatures of particular DELPH-IN tools do or don't work when using RMRS in Typed Feature Structures
- Depending on the answers to the above, how do we actually write down RMRS in TFS using TDL?
- Is there any motivation to construct an MRS representation which doesn't enforce label uniqueness (perhaps by passing around eNs rather than labels as predicate arguments)?
=== Theoretical ===
- What's the *semantic* difference between an "e" and an "x"? Is a tacit assumption about the relationship of morphosyntax to semantics taken when saying that words of a given morphosyntactic class introduce xs vs introducing es.
- Given different classes of intransitive roots (unergatives whose single arg is something like a proto-agent, and unacccusatives whose single arg is something like proto-patient) does ARGN always map to the same proto-role type or is interpretation verb-stem dependent? If the former, ARGN seems like a shorthand for the proto-role-labels, if the latter ARGN starts to seem like a stand-in for the grammatical relation of SUBJ vs OBJ (or COMPS.FIRST).
- ..
In addition to the "patient-oriented" roots discussed above, Lushootseed has agent-oriented roots. These can be used a base for deriving transitives with applicative morphology. Transitives derived from either type of root can only take a single full NP argument (termed the "direct argument" in Salishan studies), Lushootseed is notable for completely disallowing two direct NP arguments even for transitives. Thus:
- one argument or the other of a transitive is either null marked or marked via morphology (object-marking suffixes) or marked via a subject-clitic (the čəd seen in the examples above is such a clitic)
Some further facts of the system are that
- third person singular is null in the subject-clitic series and in the object-marking suffix series.
- if there is a full NP argument, and there is no object marking, the NP argument is interpreted as the more PATIENT-like of two args
- if there is a full NP argument, and there is object marking, the NP argument is interpreted as the more AGENT-like of the two args
- passivization reverses the interpretation
Thus there are no sentences such as (two full NPs, even in a transitive verb):
*ʔuʔux̌ʷ-c ti č̓ač̓as tsi č̓ač̓as (intended: the boy went after the girl)
But there are (subject marked with clitic):
ʔuʔux̌ʷ-c čəxʷ ti č̓ač̓as I went after the girl
And (object marked with morphology):
ʔuʔux̌ʷ-c-bš ti č̓ač̓as The girl went after me
Here are some further examples from Hess which show something of how causatives and applicatives are built from a couple of underlying root classes.
From Hess (1973, 89--90)
ʔuʔux̌ʷ ti č̓ač̓as
The boy went
ʔugʷədil ti č̓ač̓as
The boy sat.
--
ʔuǰiq̓ ti č̓ač̓as
The boy drowned.
ʔubəč ti č̓ač̓as
The boy fell.
ʔuč̓axʷ ti č̓ač̓as
The boy got hit (by a branch while going through thick brush).
--
ʔuʔux̌ʷ-c
He went after someone.
ʔugʷədil-s
He sat next to someone
--
ʔuǰiq̓i-t
He immersed something.
ʔubəča-t
He set something down
--
ʔuč̓axʷa-t ti č̓ač̓as
He clubbed the boy.
ʔuʔux̌ʷ-c ti č̓ač̓as
He went after the boy.
ʔuʔux̌ʷ-c-bš
The boy went after me.
A further note is that there are quite a few more types of causatives and applicatives which aren't discussed here. The language has a robust system in its lexicon for building out types of transitives in what seems to be an obviously grammaticised way. The fact that this system stands out so readily when looking at Lushootseed verb structure is one of the principle motivations for finding a semantic representation which can capture the combinatorics exhibited here.
-
Bates, Dawn and Hess, Thom and Hilbert, Vi (1994). Lushootseed Dictionary
-
Beck, David. (2009). A taxonomy and typology of Lushootseed valency-increasing suffixes. International Journal of American Linguistics 75, 533–569.
-
Beck, David. (2000). Semantic agents, syntactic subjects, and discourse topics: How to locate Lushootseed sentences in space and time. Studies in Language 24:2, 277–317.
-
Beck, David. (1996). Transitivity and causation in Lushootseed morphology. Canadian Journal of Linguistics 41, 109–140.
-
Copestake, Ann (n.d.) Draft on website:([http://www.cl.cam.ac.uk/~aac10/papers/rmrsdraft.pdf], accessed 19 Oct 2015.
-
Copestake, Ann (2009). Invited Talk: Slacker Semantics: Why Superficiality, Dependency and Avoidance of Commitment can be the Right Way to Go. In: Proceedings of the 12th Conference of the European Chapter of the ACL (EACL 2009), pages 1-9. Athens, Greece.
-
Davidson, Donald. (1967). "The Logical Form of Action Sentences". In Nicholas Rescher (ed.), The Logic of Decision and Action. University of Pittsburgh Press, pp. 81--95.
-
Davis, Henry and Matthewson, Lisa. (2009). TITLE OF ARTICLE HERE. 'Language and Linguistics Compass' REST OF CITATION HERE
-
Dowty, David (1989). "On the Semantic Content of the Notion of 'Thematic Role'" In Gennaro Chierchia, Barbara H. Partee, and Raymond Tuner (eds.), Properties, Types and Meaning, II. pp 69--129
-
Dowty, David (1990). 1990. "Thematic Proto-Roles and Argument Selection". Language. Vol. 67, No. 3, pp. 547--619
-
Hess, Thom (1973) "Agent in a Coast Salish Language". IJAL. Volume 39, No. 2.
-
Hess, Thom. (1993) "A schema for the presentation of Lushootseed verb stems". 'American Indian Linguistics and Ethnography in Honor of Laurence C. Thompson', University of Montana Occasional Papers on Linguistics no. 10, ed. by Anthony Mattina, and Timothy Montler, 113–27. Missoula, MT: University of Montana.
-
Parsons, Parsons, Terrence (1995). Thematic Relations and Arguments. Linguistic Inquiry. Vol. 26, No. 4 (Autumn, 1995), pp. 635-662
Home | Forum | Discussions | Events