Word predictability effects are linear, not logarithmic: Implications for probabilistic models of sentence comprehension

Brothers, Trevor; Kuperberg, Gina R.

doi:10.1016/j.jml.2020.104174

Cited by 57 publications

(101 citation statements)

References 69 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Our finding that both logistic and linear functions of probability predict N400 amplitudes speaks to a recent discussion in the literature about the relationship between predictability and indices of context-based facilitation (Brothers & Kuperberg, 2021;Smith & Levy, 2013; see also Yan et al, 2017). According to the proportional pre-activation account, lexical features of upcoming words get hierarchically (pre)activated from higher cortical representations, in line with a hierarchical generative framework described by Kuperberg and colleagues (2020).…”

Section: The Effect Of Contextual Facilitation Is Both Linear and Logarithmicsupporting

confidence: 86%

“…Thus, on this account, context effects should extend to all words in the lexicon and should bear a logarithmic-shaped relationship to word probability. Alternative accounts (see discussion in Brothers & Kuperberg, 2021) instead hypothesize that the comprehension system allocates activation to likely next words in proportion to their probability, creating a linear relationship between word probability and contextual facilitation. On this account, then, contextual facilitation is effectively limited in scope to words that are relatively probable in the input.…”

Section: Discussionmentioning

confidence: 99%

See 1 more Smart Citation

Context-Based Facilitation of Semantic Access Follows Both Logarithmic and Linear Functions of Stimulus Probability

Szewczyk¹,

Federmeier²

2021

Preprint

View full text Add to dashboard Cite

Stimuli are easier to process when the preceding context (e.g., a sentence, in the case of a word) makes them predictable. However, it remains unclear whether context-based facilitation arises due to predictive preactivation of a limited set of relatively probable upcoming stimuli (with facilitation then linearly related to probability) or, instead, arises because the system maintains and updates a probability distribution across all items, as posited by accounts (e.g., surprisal theory) assuming a logarithmic function between predictability and processing effort. To adjudicate between these accounts, we measured the N400 component, an index of semantic access, evoked by sentence-final words of varying probability, including unpredictable words, which are never generated in human production norms. Word predictability was measured using both cloze probabilities and a state-of-the-art machine learning language model (GPT-2). We reanalyzed five datasets (n=138) to first demonstrate and then replicate that context-based facilitation on the N400 is graded and dissociates even among words with cloze probabilities at or near 0, as a function of very small differences in model-estimated predictability. Furthermore, we established that the relationship between word predictability and context-based facilitation on the N400 is neither purely linear nor purely logarithmic but instead combines both functions. We argue that such a composite function reveals properties of the mapping between words and semantic features and how feature- and word- related information is activated during on-line processing. Overall, the results provide powerful evidence for the role of internal models in shaping how the brain apprehends incoming stimulus information.

show abstract

Section: The Effect Of Contextual Facilitation Is Both Linear and Logarithmicsupporting

confidence: 86%

Section: Discussionmentioning

confidence: 99%

Context-Based Facilitation of Semantic Access Follows Both Logarithmic and Linear Functions of Stimulus Probability

Szewczyk¹,

Federmeier²

2021

Preprint

View full text Add to dashboard Cite

show abstract

“…Note that the use of the term “prediction error” here does not imply a “prediction violation” or “error” in the colloquial sense: The first‐level prediction error produced by an incoming event simply reflects the new information provided by observing this event—the information that was not reconstructed before it was observed (see also footnote 5). The magnitude of the prediction error produced by the incoming event is proportional to the likelihood of its prior pre‐activation (see Brothers & Kuperberg, 2020), and it reflects the amount of “work” required to initially retrieve/access this event representation. Neurophysiologically, the difficulty of retrieving this event information, and therefore the magnitude of the first‐level prediction error, is thought to be reflected by the amplitude of the N400 ERP component: The more predictable the event, the smaller (less negative) the N400 (see Kuperberg, 2016; Kuperberg et al, 2020 for discussion), although note that the N400 produced in response to visual events has a more anterior scalp distribution and a more extended time course than the N400 produced by single words during language comprehension (e.g., Coderre et al, 2020; Cohn et al, 2012; Sitnikova et al, 2008; West & Holcomb, 2002).…”

Section: A Hierarchical Generative Frameworkmentioning

confidence: 99%

Tea With Milk? A Hierarchical Generative Framework of Sequential Event Comprehension

Kuperberg

2020

Topics in Cognitive Science

Self Cite

View full text Add to dashboard Cite

To make sense of the world around us, we must be able to segment a continual stream of sensory inputs into discrete events. In this review, I propose that in order to comprehend events, we engage hierarchical generative models that “reverse engineer” the intentions of other agents as they produce sequential action in real time. By generating probabilistic predictions for upcoming events, generative models ensure that we are able to keep up with the rapid pace at which perceptual inputs unfold. By tracking our certainty about other agents' goals and the magnitude of prediction errors at multiple temporal scales, generative models enable us to detect event boundaries by inferring when a goal has changed. Moreover, by adapting flexibly to the broader dynamics of the environment and our own comprehension goals, generative models allow us to optimally allocate limited resources. Finally, I argue that we use generative models not only to comprehend events but also to produce events (carry out goal‐relevant sequential action) and to continually learn about new events from our surroundings. Taken together, this hierarchical generative framework provides new insights into how the human brain processes events so effortlessly while highlighting the fundamental links between event comprehension, production, and learning.

show abstract

“…This is because the failure may be driven by many factors that are external to the research question under investigation: differences in the population and/or language studied, the natural variability in the dependent variable, lab settings, equipment, and protocols can come together to lead to very different outcomes. Indeed, it is possible that when it comes to studying subtle and highly variable aspects of human (IR)REPRODUCIBILITY IN PSYCHOLOGY AND PSYCHOLINGUISTICS (IR)REPRODUCIBILITY IN PSYCHOLOGY AND PSYCHOLINGUISTICS (2020), Brandt et al (2020), Brewer et al (2021), Bristol and Rossano (2020), Brothers and Kuperberg (2021), Brysbaert (2019), Bürki et al (2020), Chan et al (2020), Chetail (2020), Collins et al (2020), Corps and Rabagliati (2020), Dıéez-Álamo et al (2020), Falandays et al (2020), Fellman et al (2020), Floccia et al (2020), Fox et al (2020), Fujita and Cunnings (2020), Gagné et al (2020), Garnham et al (2020), Günther, Nguyen, et al (2020), Günther, Petilli, et al (2020), Hesse and Benz (2020), Hollis (2020), Humphreys et al (2020), Hwang and Shin (2019), Isarida et al (2020), Jäger et al (2020), Johns et al (2020), Kaula and Henson (2020), Lauro et al (2020), Lelonkiewicz et al (2020), Li et al (2020), Liang et al (2021), McKinley and Benjamin (2020),…”

Section: Introductionmentioning

confidence: 99%

Share the code, not just the data: A case study of the reproducibility of articles published in the Journal of Memory and Language under the open data policy

Laurinavichyute¹,

Vasishth²

2021

Preprint

View full text Add to dashboard Cite

In 2019 the Journal of Memory and Language instituted an open data and code policy; this policy requires that, as a rule, code and data be released at the latest upon publication. Does this policy lead to reproducible results? We looked at whether 57 papers published between 2019 and 2021 were reproducible, in the sense that the published summary statistics should be possible to regenerate given the data, and given the code, when code was provided. We found that for 10 out of the 57 papers, data sets were inaccessible; 29 of the remaining 47 papers provided code, of which 16 were reproducible. Of the 18 papers that did not provide code, one was reproducible. Overall, the reproducibility rate was about 30%. This estimate is similar to the ones reported for psychology, economics, and other areas, but it is probably possible to do better. We provide some suggestions on how reproducibility can be improvedin future work.

show abstract

Word predictability effects are linear, not logarithmic: Implications for probabilistic models of sentence comprehension

Cited by 57 publications

References 69 publications

Context-Based Facilitation of Semantic Access Follows Both Logarithmic and Linear Functions of Stimulus Probability

Context-Based Facilitation of Semantic Access Follows Both Logarithmic and Linear Functions of Stimulus Probability

Tea With Milk? A Hierarchical Generative Framework of Sequential Event Comprehension

Share the code, not just the data: A case study of the reproducibility of articles published in the Journal of Memory and Language under the open data policy

Contact Info

Product

Resources

About