N -gram probability effects in a cloze task (original) (raw)
Related papers
Usage-Based Individual Differences in the Probabilistic Processing of Multi-Word Sequences
2021
While it is widely acknowledged that both predictive expectations and retrodictive integration influence language processing, the individual differences that affect these two processes and the best metrics for observing them have yet to be fully described. The present study aims to contribute to the debate by investigating the extent to which experienced-based variables modulate the processing of word pairs (bigrams). Specifically, we investigate how age and reading experience correlate with lexical anticipation and integration, and how this effect can be captured by the metrics of forward and backward transition probability (TP). Participants read more and less strongly associated bigrams, paired to control for known lexical covariates such as bigram frequency and meaning (i.e., absolute control, total control, absolute silence, total silence) in a self-paced reading (SPR) task. They additionally completed assessments of exposure to print text (Author Recognition Test, Shipley vocabulary assessment, Words that Go Together task) and provided their age. Results show that both older age and lesser reading experience individually correlate with stronger TP effects. Moreover, TP effects differ across the spillover region (the two words following the noun in the bigram).
PLOS ONE, 2021
During reading or listening, people can generate predictions about the lexical and morphosyntactic properties of upcoming input based on available context. Psycholinguistic experiments that study predictability or control for it conventionally rely on a human-based approach and estimate predictability via the cloze task. Our study investigated an alternative corpus-based approach for estimating predictability via language predictability models. We obtained cloze and corpus-based probabilities for all words in 144 Russian sentences, correlated the two measures, and found a strong correlation between them. Importantly, we estimated how much variance in eye movements registered while reading the same sentences was explained by each of the two probabilities and whether the two probabilities explain the same variance. Along with lexical predictability (the activation of a particular word form), we analyzed morphosyntactic predictability (the activation of morphological features of words)...
Predicting word ’predictability’ in cloze completion, electroencephalographic and eye movement data
Previous neurocognitive approaches to word predictability from sentence context in electroencephalographic (EEG) and eye movement (EM) data relied on cloze completion probability (CCP) data effortly collected from up to 100 human participants. Here we test whether two well-established techniques in computational linguistics can predict these data. Together with baseline pre- dictors of word position and frequency, we found that n-gram language models but not topic models provide an approach to EEG and EM data that is not significantly inferior to the CCP-based predictability data. This is the case for the three corpora we used. Most strikingly, our models accounted for about half of the variance of the CCP-based predictability estimates, thus suggesting that it provides a computational framework to explain the predictability of a word from sentence context. This can help to generalize neurocognitive models to all possible novel word combinations.
Language-users choose short words in predictive contexts in an artificial language task
2017
Zipf (1935) observed that word length is inversely proportional to word frequency in the lexicon. He hypothesised that this cross-linguistically universal feature was due to the Principle of Least Effort: language-users align form-meaning mappings in such a way that the lexicon is optimally coded for efficient information transfer. However, word frequency is not the only reliable predictor of word length: Piantadosi, Tily, and Gibson (2011) show that a word’s predictability in context is in fact more strongly correlated with word length than word frequency. Here, we present an artificial language learning study aimed at investigating the mechanisms that could give rise to such a distribution at the level of the lexicon. We find that participants are more likely to use an ambiguous short form in predictive contexts, and distinct long forms in surprising contexts, only when they are subject to the competing pressures to communicate accurately and efficiently. These results support the...
Experiments on predictability of word in context and information rate in natural language.
Based on data from a large-scale experiment with human subjects, we conclude that the logarithm of probability to guess a word in context (unpredictability) depends linearly on the word length. This result holds both for poetry and prose, even though with prose, the subjects don't know the length of the omitted word. We hypothesize that this effect reflects a tendency of natural language to have an even information rate.
A model for evidence accumulation in the lexical decision task
Cognitive Psychology, 2004
We present a new model for lexical decision, REM-LD, that is based on REM theory (e.g., Shiffrin & Steyvers, 1997). REM-LD uses a principled (i.e., Bayes’ rule) decision process that simultaneously considers the diagnosticity of the evidence for the ‘WORD’ response and the ‘NONWORD’ response. The model calculates the odds ratio that the presented stimulus is a word or a nonword by averaging likelihood ratios for lexical entries from a small neighborhood of similar words. We report two experiments that used a signal-to-respond paradigm to obtain information about the time course of lexical processing. Experiment 1 verified the prediction of the model that the frequency of the word stimuli affects performance for nonword stimuli. Experiment 2 was done to study the effects of nonword lexicality, word frequency, and repetition priming and to demonstrate how REM-LD can account for the observed results. We discuss how REM-LD could be extended to account for effects of phonology such as the pseudohomophone effect, and how REM-LD can predict response times in the traditional ‘respond-when-ready’ paradigm.
A New Perspective on Word Association: How keystroke logging informs strength of word association
WORD, 2018
For many years, word association (WA) data has informed theories of the mental lexicon by analyzing the words elicited. However, findings are inconsistent and WA research is still waiting for 'a breakthrough in methodology which can unlock its undoubted potential' (Schmitt 2010: 248). In this paper, we offer a new perspective on WA by using keystroke logging (Inputlog, Leijten & Van Waes 2013) to captures the processes of word production. More specifically, we analyse pause behaviour during a continued, typed, word association task with 30 cue words eliciting 4 responses, per cue, to evaluate the strength of links in lexical selection processes. We show a strong positive correlation between pause length and inter-response location, providing empirical evidence which supports the established hypothesis that as more responses are elicited, links between them become weaker. Furthermore, using Fitzpatrick's response classification (2007), we found meaning-based responses were most common in the dataset generally, but, they particularly occurred after longer pauses, and exclusively so after the longest pauses. Position and form-based responses, whilst less frequent overall, typically followed the shortest pauses. In our conclusion we highlight the importance of our methodology in fine-tuning ongoing understanding of how we access the mental lexicon.