Stevenson and Merlo (1997) report additional difficulty in processing garden path sentences when the locally ambiguous verb is unergative, in contrast to garden path sentences where the ambiguous token is unaccusative, as demonstrated in 1.
1a. The horse raced past the barn fell. 1b. The cookies baked in the oven were tasty.
Difficulty seems to arise from an initial active intransitive parse of the verbs 'raced' and 'baked', but integrating the disambiguating tokens 'fell' or 'were tasty' requires reanalyzing the ambiguous token as passive, and therefore, transitive. To explain the asymmetry of 1a./1b., Stevenson and Merlo citing Hale and Keyser 1993, argue that unaccusatives such as 1b. 'baked' apply lexical causation, but unergatives like 1a 'raced' must apply causation syntactically.
However, models with enhanced thematic granularity also capture the pertinent causative asymmetry, obviating the need to stipulate asymmetrical costs of syntactic and lexical causation. Dowty 1991, in Appendix 1, provides relevant criteria for canonical Agent and Theme theta-roles. By adopting the Dowtyian model, we explain relative late parsing of the causative sense of "raced" in 1a; at this token, "horse" is uniformly preferred as Agent, but reanalysis yields a proto-Agent divide between the Actor-Racer "horse" and the oblique Causer. However, 1b. exhibits no such reanalysis: "cookies" are a canonical Theme at both the ambiguous token "baked" and at the disambiguating "fell".
Here, we apply the Entropy Reduction Hypothesis (Hale 2006) to a sufficiently fine-grained grammar, subsuming the Stevenson/Merlo account: performance on interesting verb tokens depends on integrative success of Dowtyian features in the manipulated garden-path presentation. Our model, containing a PCFG with coefficents of covert morphemes estimated from British National Corpus hand-tagged alternation, predicts difficulty across a continuum of potentially transitive verb classes variant in degree of protoAgent integration, of which unaccusatives and unergatives constitute the extremes, and extending to psych, control, and raising verbs.
Several accounts, (McKoon and Radcliffe 2003, Stevenson and Merlo 1997), draw critical distinctions between causative alternation of unaccusatives, which is productive, and that of unergatives, which is seen to be "extralexical"/ non-productive. Crucially, no work has, until now, controlled for the association between unaccusative verb class and internal argument animacy, and we offer that high animacy requirements for both Causer and Actor theta-roles engender comparably fewer tokens in corpus, than in unaccusatives, while Agent and Theme are maximally distinct.
An acceptability study controlled for these thematic-range effects by selecting 10 unaccusative verbs from Levin's EVCA class 45.4 and 10 unergatives from EVCA 51.3.2: agentive verbs of manner of motion, finding control for animacy in internal arguments improved acceptability of transitive unergatives (2.27; 1 acceptable, 5 unacceptable) to that of unaccusatives (2.378947). Each verb appeared in both an intransitive and transitive condition, controlled for syntactic frame: for the first, the form DP_2-Verb-PP, where DP_2 consisted of determiner "the" and animate noun; for the latter, the form DP_1-Verb-DP_2-PP, with DP_1 as a proper noun, DP_2 and PP identical to the intransitive presentation.
I'm somewhat lacking the background for understanding your work, but anyway, it sounds very interesting and reads well.
Personally I like abstracts that have a brief introduction (one sentence) that tells what the story is about. Also, for my taste, your text ends a bit abrupt and rather technical. Perhaps you could add a sentence that states your main result in simple terms.
"However, 1b. exhibits no such reanalysis: "cookies" are a canonical Theme at both the ambiguous token "baked" and at the disambiguating "fell"."
Shouldn't it be "at the disambiguating "were tasty""?
I agree, if there is some way to compress the re-telling of Stevenson and Merlo into one sentence and more quickly get to the Reason to Accept This Abstract (sentence #1 of para 3) that would be superior.
maybe something like "We present an improved model of the alternation first characterized by S and M 97...
this model is more accurate in that - it handles more real data (the data is shown to be real by an acceptability study that blah blah blah...)
this model is more parsimonious in that - it does lexical semantics in the syntax with head-movement
- use the word "model" to get into the formal models session - consider using a picture to hint that your analysis involves two covert morpheme presence/absence dimensions. I would think that would be easy to get on-board with. - avoid subjective words like "interesting", "crucially" - just tell reviewers what the result is, not how you got it (unless deviation from std methodology is part of the result)
3 comments:
Stevenson and Merlo (1997) report additional difficulty in processing garden path sentences when the locally ambiguous verb is unergative, in contrast to garden path sentences where the ambiguous token is unaccusative, as demonstrated in 1.
1a. The horse raced past the barn fell.
1b. The cookies baked in the oven were tasty.
Difficulty seems to arise from an initial active intransitive parse of the verbs 'raced' and 'baked', but integrating the disambiguating tokens 'fell' or 'were tasty' requires reanalyzing the ambiguous token as passive, and therefore, transitive. To explain the asymmetry of 1a./1b., Stevenson and Merlo citing Hale and Keyser 1993, argue that unaccusatives such as 1b. 'baked' apply lexical causation, but unergatives like 1a 'raced' must apply causation syntactically.
However, models with enhanced thematic granularity also capture the pertinent causative asymmetry, obviating the need to stipulate asymmetrical costs of syntactic and lexical causation. Dowty 1991, in Appendix 1, provides relevant criteria for canonical Agent and Theme theta-roles. By adopting the Dowtyian model, we explain relative late parsing of the causative sense of "raced" in 1a; at this token, "horse" is uniformly preferred as Agent, but reanalysis yields a proto-Agent divide between the Actor-Racer "horse" and the oblique Causer. However, 1b. exhibits no such reanalysis: "cookies" are a canonical Theme at both the ambiguous token "baked" and at the disambiguating "fell".
Here, we apply the Entropy Reduction Hypothesis (Hale 2006) to a sufficiently fine-grained grammar, subsuming the Stevenson/Merlo account: performance on interesting verb tokens depends on integrative success of Dowtyian features in the manipulated garden-path presentation. Our model, containing a PCFG with coefficents of covert morphemes estimated from British National Corpus hand-tagged alternation, predicts difficulty across a continuum of potentially transitive verb classes variant in degree of protoAgent integration, of which unaccusatives and unergatives constitute the extremes, and extending to psych, control, and raising verbs.
Several accounts, (McKoon and Radcliffe 2003, Stevenson and Merlo 1997), draw critical distinctions between causative alternation of unaccusatives, which is productive, and that of unergatives, which is seen to be "extralexical"/ non-productive. Crucially, no work has, until now, controlled for the association between unaccusative verb class and internal argument animacy, and we offer that high animacy requirements for both Causer and Actor theta-roles engender comparably fewer tokens in corpus, than in unaccusatives, while Agent and Theme are maximally distinct.
An acceptability study controlled for these thematic-range effects by selecting 10 unaccusative verbs from Levin's EVCA class 45.4 and 10 unergatives from EVCA 51.3.2: agentive verbs of manner of motion, finding control for animacy in internal arguments improved acceptability of transitive unergatives (2.27; 1 acceptable, 5 unacceptable) to that of unaccusatives (2.378947). Each verb appeared in both an intransitive and transitive condition, controlled for syntactic frame: for the first, the form DP_2-Verb-PP, where DP_2 consisted of determiner "the" and animate noun; for the latter, the form DP_1-Verb-DP_2-PP, with DP_1 as a proper noun, DP_2 and PP identical to the intransitive presentation.
Hi Kyle and John!
I'm somewhat lacking the background for understanding your work, but anyway, it sounds very interesting and reads well.
Personally I like abstracts that have a brief introduction (one sentence) that tells what the story is about. Also, for my taste, your text ends a bit abrupt and rather technical. Perhaps you could add a sentence that states your main result in simple terms.
"However, 1b. exhibits no such reanalysis: "cookies" are a canonical Theme at both the ambiguous token "baked" and at the disambiguating "fell"."
Shouldn't it be "at the disambiguating "were tasty""?
I agree, if there is some way to compress the re-telling of Stevenson and Merlo into one sentence and more quickly get to the Reason to Accept This Abstract (sentence #1 of para 3) that would be superior.
maybe something like "We present an improved model of the alternation first characterized by S and M 97...
this model is more accurate in that
- it handles more real data
(the data is shown to be real by an acceptability study that blah blah blah...)
this model is more parsimonious in that
- it does lexical semantics in the syntax with head-movement
- use the word "model" to get into the formal models session
- consider using a picture to hint that your analysis involves two covert morpheme presence/absence dimensions. I would think that would be easy to get on-board with.
- avoid subjective words like "interesting", "crucially"
- just tell reviewers what the result is, not how you got it (unless deviation from std methodology is part of the result)
Post a Comment