Advanced

The role of segmental and suprasegmental information in word-level and sentence-level predictions

Söderström, Pelle LU ; Horne, Merle LU and Roll, Mikael LU (2018) Bridging Attention and Prediction 2018
Abstract
Introduction
This contribution presents recent research results on segmental and
suprasegmental cues to word-internal and sentence-level structure. In
Swedish, prosody is tightly connected to morphological and syntactic
structure. For example, every word stem is pronounced with either a low or a
high tone. The stem tone depends on what follows the stem. Thus, if a
singular suffix is connected to the word stem (as in båt-en, ‘boat-SG
DEF’), the stem is associated with a low tone, while plural suffixes induce
a high tone onto the word stem (båt-ar, ‘boat PL INDEF). Furthermore, the
high tone also cues compound word structures. Therefore, high stem tones
consistently cue more possible continuations as... (More)
Introduction
This contribution presents recent research results on segmental and
suprasegmental cues to word-internal and sentence-level structure. In
Swedish, prosody is tightly connected to morphological and syntactic
structure. For example, every word stem is pronounced with either a low or a
high tone. The stem tone depends on what follows the stem. Thus, if a
singular suffix is connected to the word stem (as in båt-en, ‘boat-SG
DEF’), the stem is associated with a low tone, while plural suffixes induce
a high tone onto the word stem (båt-ar, ‘boat PL INDEF). Furthermore, the
high tone also cues compound word structures. Therefore, high stem tones
consistently cue more possible continuations as compared to low stem tones.
Similarly, on the sentence level, high tones at the beginning of clauses can
cue different kinds of structures (statements, questions) as opposed to low
clause-initial tones which only cue subordinate clause structure. A number of EEG and fMRI experiments have investigated the neural response to these cues, focusing on the finding that more predictively useful cues in Swedish have been connected to an ERP component, the ‘pre-activation negativity’
(PrAN), which has been found to be modulated by predictive certainty as
regards possible word or sentence continuations. Specifically, PrAN has been
found to increase in amplitude as an inverse function of the number of
possible continuations of a word (Söderström et al. 2016, Roll et al.
2017).

Methods
In word-level experiments, participants listened to sentences with nouns
carrying low or high stem tones cueing either singular or plural suffixes and
were asked to judge whether the word was singular or plural. In
sentence-level experiments, participants were asked to judge the word order
(main or subordinate clause structure) of clauses preceded by clause-initial
predictive tonal cues. ERPs and event-related fMRI have been time-locked to
both predictive cue onset (i.e. the tone) and onset of the predicted
structure. Similarly, the impact of segmental cues to upcoming word structure – the initial phonemes of a word – has also been investigated using both
fMRI and ERPs.

Results
More predictive cues at the word and sentence level appear to elicit a
left-lateralised and anterior pre-activation negativity in the ERPs, at
around 100 ms after cue onset. Word-level predictive cues, both at the
segmental and suprasegmental levels, increase activity mainly in the left
inferior parietal lobe and temporal lobe, while more predictive cues at the
sentence level correlate with activity in left inferior frontal gyrus (BA44)
and anterior insula. The amplitude of PrAN has also been found to correlate
with subsequent behavioural measures, such as response accuracy and response time. Furthermore, disconfirmed predictions lead to P600 effects, and at the word level, unexpected suffixes have given rise to either LAN or N400 effects.

Conclusions
Predictive cues – both segmental and suprasegmental – in Swedish give
rise to a left anterior ERP negativity, which increases in amplitude as a
function of (certain measures of) predictive certainty. We suggest that the
pre-activation negativity is an index of a mechanism in some way related to
the pre-activation of strongly expected linguistic information at both the
word and sentence level. (Less)
Please use this url to cite or link to this publication:
author
organization
publishing date
type
Contribution to conference
publication status
published
subject
keywords
prediction, pre-activation, prosody, ERP, fMRI, psycholinguistics, syntax, morphology
conference name
Bridging Attention and Prediction 2018
language
English
LU publication?
yes
id
965e0280-5df2-4a76-b5ce-871047c129c2
date added to LUP
2018-02-22 13:52:04
date last changed
2018-05-29 09:36:06
@misc{965e0280-5df2-4a76-b5ce-871047c129c2,
  abstract     = {Introduction<br/>This contribution presents recent research results on segmental and<br/>suprasegmental cues to word-internal and sentence-level structure. In<br/>Swedish, prosody is tightly connected to morphological and syntactic<br/>structure. For example, every word stem is pronounced with either a low or a<br/>high tone. The stem tone depends on what follows the stem. Thus, if a<br/>singular suffix is connected to the word stem (as in båt-en, ‘boat-SG<br/>DEF’), the stem is associated with a low tone, while plural suffixes induce<br/>a high tone onto the word stem (båt-ar, ‘boat PL INDEF). Furthermore, the<br/>high tone also cues compound word structures. Therefore, high stem tones<br/>consistently cue more possible continuations as compared to low stem tones.<br/>Similarly, on the sentence level, high tones at the beginning of clauses can<br/>cue different kinds of structures (statements, questions) as opposed to low<br/>clause-initial tones which only cue subordinate clause structure. A number of EEG and fMRI experiments have investigated the neural response to these cues, focusing on the finding that more predictively useful cues in Swedish have been connected to an ERP component, the ‘pre-activation negativity’<br/>(PrAN), which has been found to be modulated by predictive certainty as<br/>regards possible word or sentence continuations. Specifically, PrAN has been<br/>found to increase in amplitude as an inverse function of the number of<br/>possible continuations of a word (Söderström et al. 2016, Roll et al.<br/>2017).<br/><br/>Methods<br/>In word-level experiments, participants listened to sentences with nouns<br/>carrying low or high stem tones cueing either singular or plural suffixes and<br/>were asked to judge whether the word was singular or plural. In<br/>sentence-level experiments, participants were asked to judge the word order<br/>(main or subordinate clause structure) of clauses preceded by clause-initial<br/>predictive tonal cues. ERPs and event-related fMRI have been time-locked to<br/>both predictive cue onset (i.e. the tone) and onset of the predicted<br/>structure. Similarly, the impact of segmental cues to upcoming word structure – the initial phonemes of a word – has also been investigated using both<br/>fMRI and ERPs.<br/><br/>Results<br/>More predictive cues at the word and sentence level appear to elicit a<br/>left-lateralised and anterior pre-activation negativity in the ERPs, at<br/>around 100 ms after cue onset. Word-level predictive cues, both at the<br/>segmental and suprasegmental levels, increase activity mainly in the left<br/>inferior parietal lobe and temporal lobe, while more predictive cues at the<br/>sentence level correlate with activity in left inferior frontal gyrus (BA44)<br/>and anterior insula. The amplitude of PrAN has also been found to correlate<br/>with subsequent behavioural measures, such as response accuracy and response time.  Furthermore, disconfirmed predictions lead to P600 effects, and at the word level, unexpected suffixes have given rise to either LAN or N400 effects.<br/><br/>Conclusions<br/>Predictive cues – both segmental and suprasegmental – in Swedish give<br/>rise to a left anterior ERP negativity, which increases in amplitude as a<br/>function of (certain measures of) predictive certainty. We suggest that the<br/>pre-activation negativity is an index of a mechanism in some way related to<br/>the pre-activation of strongly expected linguistic information at both the<br/>word and sentence level.},
  author       = {Söderström, Pelle and Horne, Merle and Roll, Mikael},
  keyword      = {prediction,pre-activation,prosody,ERP,fMRI,psycholinguistics,syntax,morphology},
  language     = {eng},
  month        = {02},
  title        = {The role of segmental and suprasegmental information in word-level and sentence-level predictions},
  year         = {2018},
}