Processing frequent multi-words expressions: behavioral and electrophysiological perspectives.

Sunday, October 02nd,   2011 [11:30 - 13:10]

SY_27. Processing frequent multi-words expressions: behavioral and electrophysiological perspectives

Vespignani, F. ¹ & Cacciari, C. ²

1 University ot Trento
2 University of Modena Reggio Emilia

Despite the fact that most language processing models almost ignore them, multiword expressions (MWEs) are extremely pervasive: According to Jackendoff (1995), in American English there are as many words as there are multi-word expressions, roughly around 80000. Semantic memory is in fact a repository of a variety of knowledge that includes not only word meanings and concepts but also many types of MWEs that people learn or to which they are exposed. MWEs are not only frequent but also different and span from literal word pairs to non literal language. Recently, how MWEs are represented and processed has gained an increasing interest associated with: a) how these strings formed by co-occurring lexical units are represented in the mental lexicon; b) the role of distributional regularities in language comprehension and production; c) the role of semantic expectations in modulating contextual integration at a sentential level; d) the electrophysiological correlates of processing MWEs. The aim of the Symposium is to present these topics by bringing together researchers that present behavioral and ERP evidence relevant for highlighting them. Specifically, Arnon contributes to clarifying the ways in which the frequency and predictability of four-gram literal units affect production processes. Siyanova-Chanturia et al., present behavioral and ERP results on the processing of literal multi-word frequent phrases and the processing load associated with computing them. Vespignani and Cacciari present ERP evidence questioning the idea that a unique component (the N400) is associated with the computation of highly predictable literal and idiomatic strings. Rommers et al. show ERP evidence on how semantic processing exploits contextual information to modulate the activation of expected constituents belonging to non literal expressions. Finally Katz presents reading time and ERP evidence on dynamic meaning integration processes subserving the comprehension of proverbs.

 

TALKS

SY_27.1 - Representation and processing of frequent phrases in the brain

Siyanova-Chanturia, A. ^{1, 2} , Conklin, K. ² , Kaan, E. ³ & van Heuven, W. ²

1 University of Modena and Reggio Emilia
2 University of Nottingham
3 University of Florida

Previous research suggests that frequent multiword sequences may be represented in the mental lexicon along with morphemes and words. Such studies provide evidence for a processing advantage for frequent versus infrequent phrases. While such behavioural research is informative, it tells us little about the mechanisms involved in phrasal processing above and beyond the speed of processing. In Experiment 1a, participants read frequent phrases (knife and fork), infrequent but strongly associated phrases (spoon and fork), and semantic violations (theme and fork). In Experiment 1b, participants read the same stimuli without “and”. Experiment 2 investigated the processing of frequent phrases, their reversed forms, and semantic violations in a sentence context. An N400 was observed for semantic violations. Further, frequent phrases elicited smaller N400s than associated phrases (Experiment 1a), suggesting easier semantic integration. In addition, they elicited the P300 both in (Experiment 2) and out (Experiment 1a) of context. We attribute this finding to the phenomenon of “template matching”, wherein the target sequence activates a template that matches the upcoming information. Crucially, in Experiment 1b, where items were presented without “and”, no differences were observed between frequent phrases (knife-fork) and associates (spoon-fork). This finding implies that what drives the difference between frequent phrases and associates in Experiment 1a is the phrasal status of binomials, which is why this difference disappears in Experiment 1b, where word sequences are no longer presented in their phrasal, uniquely identifiable, form. The above findings suggest that different neural correlates underlie the processing of familiar and novel language, as evidenced by increased P300s and reduced N400s for the former. Our findings are in line with the view, according to which frequent multiword sequences are characterized by a reduced processing load.

SY_27.2 - Single or multiple expectation-verification mechanisms during language processing?

Vespignani, F. ¹ & Cacciari, C. ²

1 University of Trento
2 University of Modena Reggio Emilia

Recent neurocognitive models of language comprehension attribute a crucial role to predictive and anticipatory processing based on long term memory information. Several sources of evidence in fact show that the distributional properties of language make words rather predictable within a sentence or a text: at a processing level an appropriate context can facilitate word identification, storage, syntactic and semantic integration. The N400 component is known to be sensitive to word frequency (out of context or at sentence beginning), to semantic associations with previously processed words or pictures, and to predictability. A number of experiments showed that some of these N400 modulations, classically attributed to bottom-up integrative processes, reflect message level predictive anticipatory processing.
However a specific word can be predicted based on different sources of information. From the one side, the fact that predictions based on semantic association and world knowledge trigger similar N400 effects and the generality of the inverse relation between cloze-probability level of a constituent and N400 amplitude, suggest the existence a common neural mechanism dealing with expectations and verifications during language processing. From the other side recent ERP evidence on multiword units as idioms (e.g., "cry over spilt milk") or collocational complex prepositions (e.g., "with respect to") shows a more articulated picture with patterns other than the classic N400 effects: an interplay between P3 and N400, dissociations between lexical processing negativity and N400-700. These signatures of are discussed as evidence in favor of the existence of different cognitive mechanisms involved in the processing of unexpected constituents in specific situations. This multiplicity of psychophysiological indexes questions the idea that a unique cognitive process is responsible for processing expected constituents simply by estimating the probability of co-occurrence of words.

SY_27.3 - Context-dependent semantic processing: electrophysiological evidence from idiom comprehension

Rommers, J. ¹ , Bastiaansen, M. C. ^{1, 2} & Dijkstra, T. ^{1, 2}

1 Max Planck Institute for Psycholinguistics, Nijmegen.
2 Radboud University, Donders Centre for Brain, Cognition and Behaviour, Nijmegen.

Evidence from literal sentence comprehension suggests that sufficiently constraining sentence contexts can lead to semantically specified expectations for upcoming words. In an EEG experiment, we investigated whether this semantic expectancy extends to the case of predictable words in opaque idiomatic expressions, in which the literal word meanings are unrelated to the figurative meaning of the expression as a whole (e.g., Dutch "walk against the lamp", meaning "to get caught"). Dutch participants were presented with two types of predictable sentence contexts: literal (e.g., transl. "After lunch the electrician screwed the light bulb into the...") and idiomatic (e.g., transl. "After many transactions the fraud eventually walked against the..."). In both contexts the critical word was (1) a correct and expected word (e.g., "lamp"), (2) a word that was semantically related to the expected word (e.g., "candle"), or (3) a semantically unrelated word (e.g., "fish"). Both (2) and (3) were semantic violations. In literal contexts previous findings were replicated: a graded N400 was observed, being largest for the semantically unrelated words, intermediate for semantically related words, and smallest for correct words. In contrast, in idiomatic contexts the N400s to semantically related and unrelated words were indistinguishable. These results suggest that idiomatic contexts do not lead to activation of the semantics of the predictable words. Furthermore, in idiomatic contexts only, the violations elicited a late positivity which was independent of semantic relatedness, suggesting that the semantic violations were treated as form violations instead. The results highlight the context-dependency of semantic processing and have consequences for theories of idiom comprehension.

SY_27.4 - When words of a feather flock together: The processing of proverbs

Katz, A.

The University of Western Ontario

Comprehension involves a range of processes, each of which involves the disambiguation of multiple possibilities. In the research discussed here, we use as our prototype case the processing of proverbs because proverbs often make sense both when used either literally or figuratively (i.e. as a proverb), and so can be employed to examine (a) processing differences in integrating the literal or figurative sense of a sentence into equally supportive discourse contexts and (b) the type of contextual support that facilitates integration. Offline, reading time and ERP data will be presented. The examination of meaningful longer units, such as proverbs, is problematic for ERP studies that emphasize effects taking place at single words (with, for instance N400s) because discourse contexts might begin to influence comprehension even before the critical word is considered and continue to influence across the complete statement . With respect to ERP data, we time-lock to the first word of the critical statement, a technique that arguably captures slow wave potentials that develop over sentences and reflects the ease with which the target is integrated into an ongoing text model. Using this procedure we demonstrate processing differences that are not captured in word-by-word online reading tasks, that slow cortical potentials for proverbs are more negative at the front of the head than the same sentence used literally and that contextual cues influence integration earlier than if not present. These data point to an ongoing dynamic process of meaning integration and are problematic to several theoretical perspectives on non-literal processing, including the graded salience hypothesis.

Last Minute Change

Vespignani, F. ¹ & Cacciari, C. ²

1 University of Trento
2 Dept. Of Biomedical Sciences

***Last Minute Change***

11.30 - 11.35 Symposium presentation (Cristina Cacciari)
11.35 - 11.55 Siyanova (15' talk + 5' questions)
11.55 - 12.15 Rommers (15' talk + 5' questions)
12.15 - 12.35 Vespignani (15' talk + 5' questions)
12.35 - 12.55 Katz (15' talk + 5' questions)
12.55 - 13.10 General Discussion