Publications

Abstract

It has long been recognised that phrases and sentences are organised hierarchically, but many computational models of language treat them as sequences of words without computing constituent structure. Against this background, we conducted two experiments which showed that participants interpret ambiguous noun phrases, such as second blue ball, in terms of their abstract hierarchical structure rather than their linear surface order. When a neural network model was tested on this task, it could simulate such “hierarchical” behaviour. However, when we changed the training data such that they were not entirely unambiguous anymore, the model stopped generalising in a human-like way. It did not systematically generalise to novel items, and when it was trained on ambiguous trials, it strongly favoured the linear interpretation. We argue that these models should be endowed with a bias to make generalisations over hierarchical structure in order to be cognitively adequate models of human language.

Abstract

Recent research has established that cortical activity “tracks” the presentation rate of syntactic phrases in continuous speech, even though phrases are abstract units that do not have direct correlates in the acoustic signal. We investigated whether cortical tracking of phrase structures is modulated by the extent to which these structures compositionally determine meaning. To this end, we recorded electroencephalography (EEG) of 38 native speakers who listened to naturally spoken Dutch stimuli in different conditions, which parametrically modulated the degree to which syntactic structure and lexical semantics determine sentence meaning. Tracking was quantified through mutual information between the EEG data and either the speech envelopes or abstract annotations of syntax, all of which were filtered in the frequency band corresponding to the presentation rate of phrases (1.1–2.1 Hz). Overall, these mutual information analyses showed stronger tracking of phrases in regular sentences than in stimuli whose lexical-syntactic content is reduced, but no consistent differences in tracking between sentences and stimuli that contain a combination of syntactic structure and lexical content. While there were no effects of compositional meaning on the degree of phrase-structure tracking, analyses of event-related potentials elicited by sentence-final words did reveal meaning-induced differences between conditions. Our findings suggest that cortical tracking of structure in sentences indexes the internal generation of this structure, a process that is modulated by the properties of its input, but not by the compositional interpretation of its output.

Abstract

Visual narratives make use of various means to convey referential and co-referential meaning, so comprehenders
must recognize that different depictions across sequential images represent the same character(s). In this study,
we investigated how the order in which different types of panels in visual sequences are presented affects how
the unfolding narrative is comprehended. Participants viewed short comic strips while their electroencephalo-
gram (EEG) was recorded. We analyzed evoked and induced EEG activity elicited by both full panels (showing a
full character) and refiner panels (showing only a zoom of that full panel), and took into account whether they
preceded or followed the panel to which they were co-referentially related (i.e., were cataphoric or anaphoric).
We found that full panels elicited both larger N300 amplitude and increased gamma-band power compared to
refiner panels. Anaphoric panels elicited a sustained negativity compared to cataphoric panels, which appeared
to be sensitive to the referential status of the anaphoric panel. In the time-frequency domain, anaphoric panels
elicited reduced 8–12 Hz alpha power and increased 45–65 Hz gamma-band power compared to cataphoric
panels. These findings are consistent with models in which the processes involved in visual narrative compre-
hension partially overlap with those in language comprehension.

Abstract

In this study, we compared the performance of a long short-term memory (LSTM) neural network to the behavior of human participants on a language task that requires hierarchically structured knowledge. We show that humans interpret ambiguous noun phrases, such as second blue ball, in line with their hierarchical constituent structure. LSTMs, instead, only do
so after unambiguous training, and they do not systematically generalize to novel items. Overall, the results of our simulations indicate that a model can behave hierarchically without relying on hierarchical constituent structure.

Abstract

Incremental comprehension of head-final constructions can reveal structural attachment preferences for ambiguous phrases. This study investigates
how temporarily ambiguous PPs are processed in Dutch verb-final constructions. In De aannemer heeft op het dakterras bespaard/gewerkt ‘The
contractor has on the roof terrace saved/worked’, the PP is locally ambiguous between attachment as argument and as adjunct. This ambiguity is
resolved by the sentence-final verb. In a self-paced reading task, we manipulated the argument/adjunct status of the PP, and its position relative to the
verb. While we found no reading-time differences between argument and
adjunct PPs, we did find that transitive verbs, for which the PP is an argument, were read more slowly than intransitive verbs, for which the PP is an adjunct. We suggest that Dutch parsers have a preference for adjunct attachment of preverbal PPs, and discuss our findings in terms of incremental
parsing models that aim to minimize costly reanalysis.

Abstract

A key challenge in understanding stories and conversations is the comprehension of ‘anaphora’, words that refer back to previously mentioned words or concepts (‘antecedents’). In psycholinguistic theories, anaphor comprehension involves the initial activation of the antecedent and its subsequent integration into the unfolding representation of the narrated event. A recent proposal suggests that these processes draw upon the brain’s recognition memory and language networks, respectively, and may be dissociable in patterns of neural oscillatory synchronization (Nieuwland & Martin, 2017). We addressed this proposal in an electroencephalogram (EEG) study with pre-registered data acquisition and analyses, using event-related potentials (ERPs) and neural oscillations. Dutch participants read two-sentence mini stories containing proper names, which were repeated or new (ease of activation) and semantically coherent or incoherent with the preceding discourse (ease of integration). Repeated names elicited lower N400 and Late Positive Component amplitude than new names, and also an increase in theta-band (4-7 Hz) synchronization, which was largest around 240-450 ms after name onset. Discourse-coherent names elicited an increase in gamma-band (60-80 Hz) synchronization compared to discourse-incoherent names. This effect was largest around 690-1000 ms after name onset and exploratory beamformer analysis suggested a left frontal source. We argue that the initial activation and subsequent discourse-level integration of referents can be dissociated with event-related EEG activity, and are associated with respectively theta- and gamma-band activity. These findings further establish the link between memory and language through neural oscillations.