3.5 Mejoramiento del proceso de conversión
3.5.3 Procedimiento de conversión al sistema dual
The results, on the whole, provide evidence against the Delayed Discourse Update hypoth- esis. We observed that the processing of Subject CT remnants (which otherwise incur a cost compared to clause-final, “local” Object CF remnants) is facilitated when the subject is marked as a CT in the matrix clause using V3 word order. If contrast-assignment in the matrix clause was delayed until encountering the CT alternative (the CTE remnant), we would have expected V2 and V3 clauses in the experimental manipulation to pattern simi- larly, as the linear position of the matrix subject and the lexical material intervening between the matrix subject and the remnant (which may influence retrieval) were kept constant in the two clause types. The reduced processing penalty for subject CT remnants following V3 clauses is explained if the processor marked the subject correlate in V3 clauses as contrastive, facilitating the pairing of the contrastive remnant and its correlate.
A novel aspect of this experiment compared to past work is that clause-initial subjects do not occur in an obviously non-canonical position. Previous work has looked at clause-initial objects in SVO languages (Kaiser and Trueswell, 2004; Weskott et al., 2011; Kristensen et al., 2014) and showed that comprehenders are sensitive to deviations from non-canonical word order and the discourse conditions that license object-fronting. Clause-initial subjects in
Remnant Type in regression-contingent subsets of the first-pass time data. The wrap-up effects occurring here might be better explored in a more targeted experimental design in the future.
Estonian are temporarily compatible with occurring in a canonical subject position (Spec- FinP). Therefore, encountering the clause-initial subject (or Nominative DP in the clause- initial position) is not expected to produce a disruption to syntactic structure-building that would force the processor to consider alternative structures. My findings here show that in determining information-structural relations, the processor is sensitive to the structure of the clause as a whole (and in the present case, to the linear placement of the verb).
While there is evidence for the computation of contrast being facilitated by V3 order, it appears that other factors like the polarity of the clause also influence CT processing. This suggests that CT-marking based on V3 order is not fully automatic and encapsulated from pragmatic and contextual factors. One explanation is that non-canonical word order (perhaps particularly in silent reading, in the absence of prosodic cues) incurs a processing penalty even when it conveys a particular information structure completely unambiguously. A pragmatic cue for contrast, such as negation, may then counter the processing difficulty introduced by parsing a non-canonical structure. This possibility is in line with previous work on the processing of non-canonical word order showing that context facilitates the comprehension of non-canonical structures (e.g. Weskott et al. 2011), although more work is needed in order to separate potential processing difficulty arising from parsing non-canonical clauses (i.e. building a syntactic representation for a less frequent structure) from slowdowns associated with deeper discourse processing, such as representing a particular constituent as contrastive and computing the associated inferences.
A general finding from the Polarity analyses was that the processing of contrast is eased when the matrix clause involves negation. This is compatible with negation activating con- trastive alternatives more easily than affirmative statements. Still, Estonian has grammatical properties that may weaken the conclusions that can be drawn from the Polarity findings. There is a possible confound arising from the nature of polarity particles in CTE and CFE in Estonian. Following negative matrix clauses, contrastive remnant ellipsis requires a positive particle and following positive matrix clauses, contrastive remnant ellipsis requires a negative particle. The positive particles for CTE (k¨ull) and CFE (vaid) are distinct, which makes
the two structures easy to distinguish from each other. The negative particle mitte is used for both CTE and CFE, which means that the two structures are only distinguished from each other by their word order. While we have seen evidence that the processor is rapidly sensitive to word order variations, having this additional cue in the form of two distinct positive particles following negative matrix clauses could still facilitate the online process- ing of contrastive remnant ellipsis. Along the same line, encountering the negative particle mitte could introduce a processing slowdown associated with determining the information- structural status (CT or CF) of the remnant. Thus, further work is needed to examine how fine-grained the processing of contrast is during incremental comprehension, and whether the parser rapidly distinguishes between CTE and CFE during incremental structure-building.
Barring potential complications with distinguishing between CTE and CFE during the reading of the remnant, let us consider what the Polarity effects could tell us about the incremental processing of the V3 clause. Considering that marking the subject as a CT is facilitated in the presence of negation, it is not clear how “immediate” the assignment of contrast is – sentential negation is encountered along with the verb, once the clause-initial subject and the adverb have altready been read. If the processor assigns contrast as rapidly as possible, we would expect the clause-initial subject to be marked as a CT once it is determined that (i) the verb does not appear in the second (V2) position, which should be licensed by a the presence of a preverbal CT, and (ii) the other preverbal constituent, the speaker-oriented adverb, cannot function as a CT. Thus, the subject could be marked as a CT before the verb is even encountered, in which case the appearance of sentential negation on the verb would not be expected to influence contrast-assignment much. One possibility is that the processor is initially briefly delaying assigning contrast to the clause- initial subject until the verb is encountered, as the subject-adverb sequence is temporarily compatible with being [-Focus, -Contrast], if followed by a CT constituent. As discussed in Chapter 2, multiple discourse-given constituents can precede a CT in the preverbal domain. Example (78) illustrates the temporary ambiguity, where the (78a) is a reproduction of the V3-SubjectCT condition from Experiment 1, and (78b) shows a temporarily string-identical
example where the subject is not a CT. The identical beginnings of the two clauses are underlined for convenience. A comparison of (78a) and (78b) shows that the information structure of the subject is not fully disambiguated until the verb has been encountered.
(78) a. AgnesCT Agnes.nom tegelikult actually tunneb knows Joonast, Joonas.part KatrinCT Katrin.nom mitte. neg
‘Actually Agnes knows Joonas, but Katrin doesn’t.’ b. Agnes Agnes.nom tegelikult actually JoonastCT Joonas.part tunneb, knows KatrinitCT Katrin.part mitte. neg
‘Joonas, Agnes actually knows, but Katrin she doesn’t.’
In Experiment 2 below, I explore how the processor deals with information-structural ambiguity in the preverbal domain in order to shed more light on how contrast is assigned in V3+ clauses during incremental processing.