|
Size: 4144
Comment:
|
Size: 4553
Comment:
|
| Deletions are marked like this. | Additions are marked like this. |
| Line 1: | Line 1: |
| #acl HlpLabGroup:read,write,delete,revert,admin | #acl HlpLabGroup,HlpLabGuestGroup:read,write,delete,revert,admin All:read |
| Line 7: | Line 7: |
| The two rapid adaptation experiments reported in Fine, Jaeger, Farmer, & Qian (submitted) provide evidence that humans rapidly adjust their subjective representation of the statistics of the input in order to more efficiently process language. The study leaves open several questions. One question is, assuming adaptation is implemented via a learning mechanism of some kind, what kind? There are numerous possibilities, but in this experiment we focus on distinguishing two specific proposals. The first proposal is the one outlined in our paper, namely that the type of learning involved in adaptation is at least computationally expressible as Bayesian belief update--each sentence that comprehenders process counts as a piece of evidence that is used to update a (subjective representation of a) probability distribution over syntactic structures. This predicts that if subjects process many main verb (MV) structures such as (1), this should take probability mass away from a competing structure such as a reduced relative clause (RC), like (2). | |
| Line 9: | Line 8: |
| (1) MV: The experienced soldiers warned about the dangers '''before the midnight''' raid. | Kaschak and Glenberg investigate how exposure to non-standard needs+participle sentences, as in (1), which are common in some American dialects, affects processing of the standard use of needs+participle, as in (2), where the participle is a modifier to the noun. |
| Line 11: | Line 10: |
| (2) RC: The experienced soldiers warned about the dangers '''conducted the midnight''' raid. | (1) The meal needs cooked given that dinner is in an hour. (non-standard needs) |
| Line 13: | Line 12: |
| (3) MV (unambiguous): The experienced soldiers spoke about the dangers '''before the midnight''' raid. | (2) The meal needs cooked vegetables to make it complete. (modifier) |
| Line 15: | Line 14: |
| (4) RC (unambiguous): The experienced soldiers who were told about the dangers '''conducted the midnight''' raid. | |
| Line 17: | Line 15: |
| Presumably this is why garden path effects are observed with sentences like (2) in the first place (subjects have far more experience with sentences like (1)). We have extremely preliminary evidence for this in Experiment 2 of Fine et al. (submitted). In that experiment, subjects see far more RCs than they expect a priori, and relatively few MV structures (few, relative to the number of RCs). By the end of the experiment, the ambiguity effect for RCs (RTs in the bold region in 2 minus RTs in the bold region of 4) had completely disappeared, whereas the ambiguity effect for MVs (RTs in the bold region of 1 minus RTs in the bold region of 3) had increased, numerically though non-significantly. | The episodic processing proposal advanced by Kaschak and Glenberg predicts that readers will initially misparse sentences such as (1) as having the more frequent modifier structure (i.e., they are garden-pathed). This retrieval should subsequently facilitate processing of that modifier structure, even though the modifier interpretation is ultimately ruled out. In several reading experiments, Kaschak and Glenberg find that repeated exposure to the temporarily ambiguous non-standard structure, compared to repeated exposure to the unambiguous standard needs to be+participle structure in (3), facilitates later processing of both the non-standard needs structure (1) and the modifier structure (2), in line with the episodic processing account. |
| Line 19: | Line 17: |
| This pattern of results is predicted by a belief-update model of learning. | (3) The meal needs to be cooked given that dinner is in an hour. (standard needs) |
| Line 21: | Line 19: |
| By contrast, Kaschak and Glenberg (2004) offer an episodic account of syntactic adaptation which predicts a different pattern of results. Specifically, under their proposal, reading a temporarily ambiguous string like ''The experienced soldiers warned about the dangers...'' causes readers to retrieve both structures, the MV and the RC. This leaves an episodic trace for '''both''' structures, which in turn facilitates processing of both. Thus, if subjects see a bunch of sentences like (2), sentences like (1) should actually get easier to process, not harder. | What does the episodic processing account say about MV/RC ambiguity? Consider a temporarily ambiguous sentence of the kind employed in the experiments reported above. |
| Line 23: | Line 21: |
| This experiment provides a more direct attempt to adjudicate between these two proposals. | (4) The experienced soldiers warned about the dangers conducted the midnight raid. |
| Line 25: | Line 23: |
| MV structures are FAR more common than RCs, so a balanced design in which MVs and RCs are interspersed (like Experiment 2 in Fine et al.) may not provide subjects with sufficient evidence against MVs to detect an actual increase in the processing cost of that structure. In order to address this, we employ a block design (below I refer to blocks as phases), which is outlined in the table below. | The episodic processing proposal advanced by Kaschak and Glenberg predicts that, in sentences such as (4), subjects will initially retrieve the MV representation at the verb warned, since there is a stronger a priori expectation for this structure than for the RC structure, and, crucially, this retrieval should subsequently facilitate processing of that structure, even though the MV interpretation is ultimately ruled out. This prediction is not supported by our experiments. On the contrary, Experiment 2 from Fine et al. finds that repeated exposure to RCs leads to slower processing of MVs (as predicted by the hypothesis advanced here). It is likely that both "episodic processing" and statistical learning are operative in rapid syntactic adaptation, but that the two processes are engaged to different degrees depending on properties of the materials used. This experiment tries to being to tease apart these two processes. Specifically, we ask what happens when subjects are exposed to unambiguous RCs only (as in (8)). If adaptation is purely a matter of subjects incrementally updating the parameters of a probability distribution over syntactic structures, then exposure to unambiguous RCs should have the same effect on the subsequent comprehension of MVs as exposure to both ambiguous and unambiguous RCs. On the other hand, if adaptation depends on episodic traces left by retrieving particular structures, then a "less jarring" structure such as an unambiguous RC should have attenuated effects on the comprehension of MVs, relative to what was found by Fine et al. (under review). (5) MV: The experienced soldiers warned about the dangers '''before the midnight''' raid. (6) RC: The experienced soldiers warned about the dangers '''conducted the midnight''' raid. (7) MV (unambiguous): The experienced soldiers spoke about the dangers '''before the midnight''' raid. (8) RC (unambiguous): The experienced soldiers who were told about the dangers '''conducted the midnight''' raid. We test this prediction in a between-subjects experiment. |
| Line 30: | Line 44: |
| || ||Phase 1||Phase 2||Phase 3|| ||Group 1||8 ambiguous RCs & 8 unambiguous RCs||5 unambiguous MVs & 5 ambiguous MVs & 20 fillers||5 unambiguous RCs & 5 ambiguous RCs & 15 fillers|| ||Group 2||16 fillers||5 unambiguous MVs & 5 ambiguous MVs & 20 fillers||5 unambiguous RCs & 5 ambiguous RCs & 15 fillers|| ||Group 3||8 ambiguous RCs & 8 unambiguous RCs||5 unambiguous RCs & 5 ambiguous RCs & 20 fillers||5 unambiguous MVs & 5 ambiguous MVs & 15 fillers|| ||Group 4||16 fillers||5 unambiguous RCs & 5 ambiguous RCs & 20 fillers||5 unambiguous MVs & 5 ambiguous MVs & 15 fillers|| |
|| ||Phase 1||Phase 2|| ||Ambiguous RC training||12 ambiguous RCs (3 verbs repeated 4x) + 15 fillers||12 MVs (6 ambiguous; 3 verbs repeated 4x) + 12 ambiguous RCs (3 verbs repeated 4x) + 24 fillers|| ||Unambiguous RC training||12 unambiguous RCs (3 verbs repeated 4x) + 15 fillers || 12 MVs (6 ambiguous; 3 verbs repeated 4x) + 12 unambiguous RCs (3 verbs 4x) + 24 fillers || To counterbalance ambiguity within groups, two lists were created for each group. Two pseudo-randomized orders were then created for each list, resulting in 4 lists per group. The materials for this experiment can be found in an excel spreadsheet here [[attachment:EpisodicStatisticalMaterials]] |
Synopsis and Design
Kaschak and Glenberg investigate how exposure to non-standard needs+participle sentences, as in (1), which are common in some American dialects, affects processing of the standard use of needs+participle, as in (2), where the participle is a modifier to the noun.
(1) The meal needs cooked given that dinner is in an hour. (non-standard needs)
(2) The meal needs cooked vegetables to make it complete. (modifier)
The episodic processing proposal advanced by Kaschak and Glenberg predicts that readers will initially misparse sentences such as (1) as having the more frequent modifier structure (i.e., they are garden-pathed). This retrieval should subsequently facilitate processing of that modifier structure, even though the modifier interpretation is ultimately ruled out. In several reading experiments, Kaschak and Glenberg find that repeated exposure to the temporarily ambiguous non-standard structure, compared to repeated exposure to the unambiguous standard needs to be+participle structure in (3), facilitates later processing of both the non-standard needs structure (1) and the modifier structure (2), in line with the episodic processing account.
(3) The meal needs to be cooked given that dinner is in an hour. (standard needs)
What does the episodic processing account say about MV/RC ambiguity? Consider a temporarily ambiguous sentence of the kind employed in the experiments reported above.
(4) The experienced soldiers warned about the dangers conducted the midnight raid.
The episodic processing proposal advanced by Kaschak and Glenberg predicts that, in sentences such as (4), subjects will initially retrieve the MV representation at the verb warned, since there is a stronger a priori expectation for this structure than for the RC structure, and, crucially, this retrieval should subsequently facilitate processing of that structure, even though the MV interpretation is ultimately ruled out. This prediction is not supported by our experiments. On the contrary, Experiment 2 from Fine et al. finds that repeated exposure to RCs leads to slower processing of MVs (as predicted by the hypothesis advanced here).
It is likely that both "episodic processing" and statistical learning are operative in rapid syntactic adaptation, but that the two processes are engaged to different degrees depending on properties of the materials used. This experiment tries to being to tease apart these two processes. Specifically, we ask what happens when subjects are exposed to unambiguous RCs only (as in (8)). If adaptation is purely a matter of subjects incrementally updating the parameters of a probability distribution over syntactic structures, then exposure to unambiguous RCs should have the same effect on the subsequent comprehension of MVs as exposure to both ambiguous and unambiguous RCs. On the other hand, if adaptation depends on episodic traces left by retrieving particular structures, then a "less jarring" structure such as an unambiguous RC should have attenuated effects on the comprehension of MVs, relative to what was found by Fine et al. (under review).
(5) MV: The experienced soldiers warned about the dangers before the midnight raid.
(6) RC: The experienced soldiers warned about the dangers conducted the midnight raid.
(7) MV (unambiguous): The experienced soldiers spoke about the dangers before the midnight raid.
(8) RC (unambiguous): The experienced soldiers who were told about the dangers conducted the midnight raid.
We test this prediction in a between-subjects experiment.
|
Phase 1 |
Phase 2 |
Ambiguous RC training |
12 ambiguous RCs (3 verbs repeated 4x) + 15 fillers |
12 MVs (6 ambiguous; 3 verbs repeated 4x) + 12 ambiguous RCs (3 verbs repeated 4x) + 24 fillers |
Unambiguous RC training |
12 unambiguous RCs (3 verbs repeated 4x) + 15 fillers |
12 MVs (6 ambiguous; 3 verbs repeated 4x) + 12 unambiguous RCs (3 verbs 4x) + 24 fillers |
To counterbalance ambiguity within groups, two lists were created for each group. Two pseudo-randomized orders were then created for each list, resulting in 4 lists per group.
The materials for this experiment can be found in an excel spreadsheet here EpisodicStatisticalMaterials