Accessibility is no alternative to alternatives
![Sample our Bioscience journals, sign in here to start your access, Latest two full volumes FREE to you for 14 days]({"type":"image" , "src" : "/sda/5925/TOC-Subject-Sample-2021-Bioscience.jpg"})
Abstract
Linguistic constituents that encode salient information are often prosodically reduced. Recent studies have presented evidence that higher contextual accessibility of referents results in lower prosodic prominence. Accounts of reduction in terms of accessibility set out to explain a range of phenomena that include those that are in the domain of linguistic theories of focus and givenness. The tacit assumption is that more general and independently motivated accessibility factors will be able to supplant the more specialised grammatical accounts of prosodic prominence. This paper reviews previous results and finds that existing accessibility accounts cannot explain a range of data easily captured by the alternatives theory of focus, and that various experimental studies motivating the accessibility view actually fail to distinguish between the two accounts. New experimental data are presented that tease the effects of accessibility and linguistic focus apart.
Acknowledgements
We would like to thank Duane Watson for his support and comments, and making available to us the stimuli of the original experiment that our study is based on. We also want to thank Jennifer Arnold for helpful and very detailed comments on earlier versions of this manuscript. Further thanks go to the anonymous reviewers for Language, Cognition and Neuroscience, and very helpful suggestions of the action editor, Ted Gibson.
Funding
This research was supported by FQRSC [grant NP-132516]: La prosodie: production, perception et différences interlinguistiques and a SSHRC/CFI Canada Research Chair in Speech and Language Processing.
Notes
1. Note that this use of the word ‘focus’ in the literature on attention is unrelated to the notion of ‘focus’ as it used in linguistic work, a notion we will define later. It is the linguistically ‘given’ material that tends to be attentionally in ‘focus’, whereas it is novel information that tends to be linguistically ‘focused’.
2. Apart from the likelihood of the accentuation status (arguably a categorical distinction), there is also evidence that prosodic prominence is gradiently adjusted depending on degrees of accessibility. Brennan (Citation1995), for example, found that pronouns are phonetically longer when their antecedents are not salient in the context, and Fowler, Levy, and Brown (Citation1997) found that the same factors that favour the choice of pronouns over proper names also have an effect on the degree of prominence among proper names.
3. Some researchers assume that there is a default location for placing prosodic prominence. See Wagner (Citation2005b) and references therein for a discussion.
4. We follow the argument in Wagner (Citation2005a, Citation2005b) that marking focus and marking givenness can both be achieved by using an alternatives-based focus operator like ∼ – as opposed to other theories that posit that there is both focus-marking and givenness-marking.
5. Rooth (Citation1992a), on the other hand, allows for instances of ∼ in which no constituent in its scope carries an F-mark.
6. We are assuming that the word square carried an accent in the target sentence even if it is not orthographically marked as such in the paper, because otherwise, the overall sentence should have sounded infelicitous, since there is no reason not to accent the word in this context. From the point of the view of the eye-tracking study, the accentual status of square should not matter, since the dependent variable was eye-movements time-locked to the offset of the word candle.
7. In fact, Terken and Hirschberg (Citation1994), a widely cited accessibility-based paper, already showed a related finding which conflicts directly with the basic expectations of thematic accessibility, to be discussed in more detail later.
8. Note that the study does not tease apart grammatical role vs. thematic role, so either interpretation would be valid.
9. The effect in the opposite direction might have phonological reasons: Post-focal prosodic reduction is much stronger than pre-focal reduction (cf. Wagner, Citation2005b, and references therein). Our experimental results reported later show a similar asymmetry based on two different thematic roles.
10. There still may be a pure phonological effect of position in addition to the focus effect. For such a phonological effect on prosodic prominence, see Wagner (Citation2012).
11. This type of construction is often referred to as ‘right-node raising’, following Ross (Citation1967).
12. I assume that just as for antecedents for pronouns that recency is an important factor for choosing between potential focus-antecedents (Clark & Sengul, Citation1979).
13. We aimed for 40 participants and accidentally ran two more than planned.
14. We used the lmer function of the lme4 R package (Bates & Maechler, Citation2010). Mixed regression models can control for participant and item random effects at the same time, and have therefore been argued to be more appropriate for statistical analyses in linguistic experimental studies than conventional analysis of variance analyses (Baayen, Davidson, & Bates, Citation2008).
15. For this and all other models reported in this paper, the contribution of the random effects was significant or close to significant, as tested by model comparisons with simpler models with one or the other random effect dropped. We included maximal random effects irrespective of whether their contribution reached significance. See Barr, Levy, Scheepers, and Tily (Citation2013) for a rationale why this is the appropriate analysis for data of this kind. Adding or dropping one of the random effects did not affect the qualitative outcome of the model.
16. For what it is worth, the model with both factors shows an effect for intensity for Position and no significant effects for the other acoustic parameters, and no effects whatsoever for Repetition.
17. Watson and Arnold (Citation2005) and Watson (Citation2010) do not report any inferential statistics beyond than error bars, which similarly suggest no effect of repetition. According to additional information provided by the authors, the effect of repetition was also not significant in the original study, although it did have an effect on the choice of a pronoun over a full noun phrase.
18. In this case, the model did not converge unless we centred the dependent variables. We generally left the dependent variables uncentred in the other models whenever the inferential statistics were virtually identical to those for the centred model, since it makes the model output easier to interpret.
19. An unpublished experiment with a similar set up as our Experiment 2 was already reported in Watson, Tanenhaus, and Gunlogson (Citation2008).
20. We ran an extra 20 participants for Experiment 2 in order to increase the power of the study and to give the repetition effect an additional chance to surface, but the results were qualitatively similar for the first 40 participants, which is the number of participants we had aimed for in Experiment 1.