Constructional contamination: How does it work and how do we measure it?

Abstract

In this article, we introduce the effect of “constructional contamination”. In constructional contamination, a subset of the instances of a target construction deviate in their realization, due to a superficial resemblance they share with instances of a contaminating construction. We claim that this contaminating effect bears testimony to the hypothesis that language users do not always execute a full parse while interpreting and producing sentences. Instead, they may rely on what has been called “shallow parsing”, i. e., chunking the utterances into large, unanalyzed exemplars that may extend across constituent borders. We propose several measures to quantify constructional contamination in corpus data. To evaluate these measures, the Dutch partitive genitive is taken under scrutiny as a target construction of constructional contamination. In this case study, it is shown that neighboring constructions play a crucial role in determining the presence or absence of the

The changing functions of competing forms: Attraction and differentiation

“…That this is so is supported by the existence of a parallel type of change. Formally similar expressions also tend to become more alike, even when they are semantically unrelated, as argued by Pijpops and Van De Velde (2016).…”

Meaningfulness Beats Frequency in Multiword Chunk Processing

“…Indeed, the prevalence of multiword sequences and their heterogeneity has famously been flagged as “a pain in the neck” (Sag, Baldwin, Bond, Copestake, & Flickinger, 2002), a major obstacle for computer science approaches to language (and the focus of special issues in computational linguistics journals: e.g., Ramisch, Villavicencio, & Kordoni, 2013; Villavicencio, Bond, Korhonen, & McCarthy, 2005). Our study suggests that going beyond distributional information to incorporate the meaningfulness of multiword sequences into corpus analyses might provide a useful way forward, not only for computational natural language processing but also for usage‐based approaches to language (for the latter, see also Pijpops & Van de Velde, 2016).…”

“…Recently, Christiansen and Chater (2016) argued that the use of such multiunit chunks is necessary to deal with the onslaught of input during real-time language acquisition and processing, given the severe memory and attentional constraints inherent to the language system (the so-called Now-or-Never bottleneck). This perspective emphasizes the role of "shallow" parsing in normal language processing (e.g., Pijpops & Van de Velde, 2016), whereby the input is chunked into larger units, and where the focus of processing is on arriving at a "good enough" interpretation of the utterance (e.g., Ferreira & Patson, 2007), rather than a full syntactic parse. Aspects of this theory were implemented in a cross-linguistic computational model capturing aspects of comprehension (shallow parsing) and production (word-chunk sequencing) in language acquisition (McCauley & Christiansen, 2019).…”

“…Our findings suggest that the meaningfulness of a multiword sequence has a direct influence on processing speed: The more a multiword sequence conveys a coherent meaning, the more likely it is that it will be processed as a linguistic unit in its own right. This may suggest that multiword sequences are stored as unanalyzed wholes (e.g., Pijpops & Van de Velde, 2016), possibly somewhat similar to single words. Indeed, a large‐scale corpus analysis by Williams et al (2015) showed that multiword sequences generally provide a better fit with a Zipfian distribution than single words (Zipf, 1935).…”

Y a -t- il une la place pour la variation dans la grammaire de construction cognitive? Le cas de la construction [por SN de X]

“…Le principe de non synonymie est le plus souvent violé, en particulier par différents types de changements diachroniques. L'analyse menée par Van de Velde (2014), par exemple, montre que tantôt la forme tantôt la signification d'une construction est susceptible d'être contaminée par une construction voisine en déclenchant des polysémies (PIJPOPS et VAN DE VELDE, 2016). De plus, des constructions formellement proches peuvent développer des similitudes fonctionnelles ou l'inverse.…”