**Poster C53**
We are happy to answer questions via Zoom, through the comments bubble in OSF (upper-right hand corner), or you can email us at shohini@umd.edu.
Under the Files section, you will find a video spiel along with a pdf version of the poster and abstract.
----------
**Introduction**
------------
It is well-known that verbs' argument structure guides language comprehension e.g. Boland (2005, 1993); Shapiro et al. (1991), Trueswell et al. (1993), McRae et al. (1998) among others. In this study, we investigate
the neural correlates of various aspects of argument structure, such as diathesis alternations, selectional restrictions, and subcategorization using contextually-situated stimuli and incorporating computational metrics to operationalize these different aspects of argument structure. Thematic fit (operationalized through selectional restrictions) and thematic ambiguity (operationalized through diathesis alternations and subcategorization entropy) is shown to correspond to different brain regions based on fMRI data during naturalistic listening.
**Methods**
-------
Participants (n=51, 32 female) listened to The Little Prince’s audiobook for 1h38min. Participants' comprehension was confirmed through
multiple-choice questions (90% accuracy, SD = 3.7%). Preprocessing was done with AFNI16 and ME-ICAv3.2 (Kundu et al. 2012). 2948 verbs in total were
identified using the NLTK toolkit and Stanford POS tagger.
Excluding modals, auxiliaries, and gerunds, there are 1970 verbs attested in the story (401 unique). Diathesis alternations for a given verb was calculated
from PropBank (Kingsbury, 2002), which consists of all the sentences from the Penn Treebank annotated with semantic roles. Higher PropBank scores indicate more diathesis alternations. Selectional restriction was calculated according to Resnik (1996) by estimating verb-direct object pairs from the Gigaword (Ferraro et al., 2014) & WaCkypedia (Baroni et al., 2009) corpora and then calculating the number of different WordNet semantic classes a given verb's direct objects falls into. Higher selectional restriction scores indicate the verb is more particular about the kinds of arguments it takes as its direct object. Apart from these semantic constraints, a third metric Subcategorization Frame (SCF) entropy operationalizes the syntactic constraint of subcategorization present in argument structure. The entropy of the subcategorization frame distribution
is a combined measure of the number of possible syntactic frames of a verb and the extent to which their distribution is balanced, reflecting the
degree of uncertainty about the syntactic constituent following the verb (Linzen et al. 2013). Higher SCF entropy scores represent higher uncertainty about the syntactic category of the verb’s complement.
We regressed the predictors described above against fMRI timecourses recorded during passive story-listening in a whole-brain analysis. In addition to
these three regressors of theoretical interest, we included in the GLM analysis (SPM12) four regressors of non-interest: time stamp of each word
offset, the log-frequency of each word in movie subtitles (Brysbaert & New 2009), the pitch (f0) and intensity (rms) of the narrator's voice.
**Results**
-------
The largest clusters for diathesis alternations were observed in
the bilateral Precuneus, the Supramarginal gyrus, MTG, MFG, & SFG on the
right and the Middle Occipital on the left. The largest clusters for
selectional restrictions were observed in the SMA, IFG Pars
Orbitalis/Triangularis, & STG. For syntactic subcategorization, largest
clusters were observed in Medial Frontal gyrus on the right and IPL on the
left. The whole-brain effects were FWE-corrected (T-score > 5.3) and shown
in Fig. 1.
**Conclusion**
----------
Our results illustrate the centrality of supramarginal gyrus,
posterior MTG & STG to processing verbs and its argument structure. This is
consistent with Thompson and Meltzer-Asscher (2014)’s neurocognitive model
of argument structure processing. They propose that once verbs are
encountered, the bilateral angular/supramarginal gyri are activated to
support retrieval of associated argument structure information. This
information, along with the initial structure, is transmitted to posterior
temporal regions (MTG and STG) for sentence-level semantic and syntactic
integration. Furthermore, for the most part our results corroborate
existing work on argument structure e.g. Ben-Shachar et al. (2003);
Shetreet et al. (2007, 2009, 2010); Thompson et al. (2007, 2010);
Meltzer-Asscher et al. (2013, 2015); den Ouden et al. (2009), and Fabre
(2017) among others. This is notable since prior neuroimaging studies were
controlled, task-based designs, and often included lexical decision tasks.
However, this study differs in that the neural bases of argument structure
was investigated in an ecologically valid setting within a naturalistic
language comprehension study and we found comparable results.
[1]: https://umd.zoom.us/my/shohini