Research

My research agenda involves the integration of computational and psycholinguistic methodologies to shed light on how our exposure to the surrounding linguistic environment shapes our language behavior. I use a wide range of approaches, including eye-tracking during reading, distributional semantic models, corpus-based models of lexical strength, lexical decision latencies and other corpus-based and behavioral techniques.

Representation and processing of multiword and idiomatic language

My research has especially focused on the representation and processing of multiword and idiomatic language. The increasing availability of corpora over the past decades has made it clear that most of our linguistic production involves fixed and recurrent multiword units like idioms, whose meaning often goes beyond the sum of their parts (e.g. break the ice, spill the beans, etc.). In this regard, understanding how they are represented and processed can give us insights about the general structure of the mental lexicon.

Previous evidence from our lab (Libben & Titone 2008; Titone et al. 2019) supports a hybrid model of idiom processing, where idioms are both directly retrieved from memory and compositionally analyzed over different time courses during processing.

In an eye-tracking study published on Memory & Cognition (Senaldi, Wei, Gullifer & Titone, 2022), my colleagues and I found that English L1-French L2 bilinguals were slower at reading idioms containing a momentary language switch as compared to switched literal controls. These data provided additional support to the role of direct retrieval in L1 idiom comprehension.

In a follow-up study appeared on Languages (Senaldi & Titone, 2022), we observed that L2 readers parse idioms compositionally, with L2 idioms being easier to process if the meaning of the single idiom components is more suggestive of the overall figurative meaning (i.e. increased verb/noun relatedness).

Of note, participants from both studies were faster at reading idioms that possessed a translational equivalent in their other language (e.g., English break the ice – French briser la glace), suggesting the intriguing possibility that the bilingual lexicon might also be integrated beyond the level of single words.

Contextual diversity as a lexical organizer of multiword units

The availability of text corpora has granted us immediate access to Big Data repositories of natural language use. From these, we can extract information on people’s linguistic behavior thanks to state-of-the-art computational language models. In a contribution for the Canadian Journal of Experimental Psychology (Senaldi, Titone & Johns, 2022), we demonstrated how the analysis of corpus-based computational measures of idioms' lexical strength can provide us with insight on the general functioning of the mental lexicon. In this study, idiom familiarity ratings turned out to be better predicted by contextual diversity measures encoding the communication patterns of Reddit users across different discourse topics (e.g., subreddits), as compared to frequency. While frequency is based on a principle of repetition, contextual diversity encodes a principle of likely need, adapted from the rational analysis of memory (Anderson, 1974), whereby phrases that are encountered across more diverse contexts are deemed as likelier to occur in future contexts, and become thus more readily accessible from memory. These results mirror previous findings for single words (Jones et al., 2012, Johns, 2021) and suggest likely need as the cognitive principle driving the lexical organization of single words and multiword strings alike.

Experiential and environmental effects on language behavior

In an ongoing project funded by the Social Sciences and Humanities Research Council of Canada (SSHRC), for which I serve as the Principal Investigator (co-PIs: Debra Titone, Brendan Johns), I am analyzing how the varying impact of large-scale real-world events (e.g., the COVID-19 pandemic, natural disasters, social movements) over time and across different contexts is reflected in the emergence of new language patterns on social media and news sources. On the behavioral side, I am using a battery of online tasks and questionnaires to understand how speakers process the language describing such events differently based on the impact of such events on their environment, and their social and linguistic background. In addition to validating usage-based language models through real-world case studies, this project will reveal how linguistic data can provide time- and place-accurate information on how societies around the globe are impacted by collectively experienced events.

Distributional semantic approaches to semantic compositionality and semantic change

In past work, I investigated issues related to semantic compositionality and semantic change through the application of vector-based distributional semantic models. Distributional models represent words and phrases as vectors encoding their co-occurrence statistics in corpora (Lenci, 2018). In a chapter of my doctoral dissertation, I proposed a vector-based index of semantic compositionality and idiomaticity that assesses the idiomatic/non-idiomatic status of a target phrase by measuring its cosine similarity with lexical variants obtained from the same phrase (Senaldi et al., 2016; 2017). In a side project, my colleagues and I explored the potential of distributional semantic models to track diachronic semantic change in the lexicon of Ancient Greek (Rodda et al., 2017).