A Probabilistic Earley Parser as a Psycholinguistic Model

By John Hale. North American Chapter of the Association for Computational Linguistics, 2001

1,461 citations recorded by the source corpus.

In human sentence processing, cognitive load can be defined many ways. This report considers a definition of cognitive load in terms of the total probability of structural options that have been disconfirmed at some point in a sentence: the surprisal of word wi given its prefix wo...i-1 on a phrase-structural language model. These loads can be efficiently calculated using a probabilistic Earley parser (Stolcke, 1995) which is interpreted as generating predictions about reading time on a word-by-word basis. Under grammatical assumptions supported by corpus-frequency data, the operation of Stolcke's probabilistic Earley parser correctly predicts processing phenomena associated with garden path structural ambiguity and with the subject/object relative asymmetry.

Subfields: Artificial Intelligence, Cognitive Neuroscience, Developmental and Educational Psychology

View Paper: https://www.semanticscholar.org/paper/a959ae0f08228d5d04eb46d5dc1ec3f4c9ee91d7

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Cognitive Neuroscience

In Developmental and Educational Psychology

Citations in the corpus

Benchmarking Hierarchical Script Knowledge (2019, background): “The sur-prisal of a word (Attneave, 1959; Hale, 2001) is its negative log probability under the model: − log ( P ( w i | w 1: i − 1 )) .”
Assessing Relative Sentence Complexity using an Incremental CCG Parser (2016, background): “For the revealing operations, we also add additional features which indicate the depth of the revealing which is analogous to sur-prisal (Hale, 2001).”
A Systematic Assessment of Syntactic Generalization in Neural Language Models (2020, background): “In probabilistic language models, these garden-path disambiguation effects are well captured by word negative log probabilities, or SURPRISALS (Hale, 2001): S(w|C) = − log2 p(w|C), which are independently well-established to predict human incremental processing difficulty over several orders of magn”
SPICE: Semantic Propositional Image Caption Evaluation (2016, background): “If this assumption is untrue in a particular application, a ﬂuency metric, such as surprisal [31,32], could be included in the evaluation.”
Investigating locality effects and surprisal in written English syntactic choice phenomena. (2016, positive): “Assuming strings of a language are generated by PCFG rules, the prefix probability of each word wk is calculated by summing the probabilities of all trees T spanning words w1 to wk: P (w1...wk) = ∑ T P (T,w1...wk) (1) Surprisal (Hale 2001) is then estimated by substituting this into the previous equ”
Speakers optimize information density through syntactic reduction (2006, background): “The diffi culty incurred by a comprehender in processing a word wi is positively correlated with its surprisal (Hale, 2001; Le vy, 2006).”
Coherence and Coreference Revisited (2007, background): “It also follows recent work in sentence processing that contends that online measurements of interpretation difficulty can be successfully predicted by probabilistic, expectation-driven models (e.g. Hale 2001; Levy 2007).”
If Beam Search Is the Answer, What Was the Question? (2020, background): “aspects of human language processing in terms of information theory; it is often applied to an area of psycholinguistics known as sentence processing where the UID hypothesis is used to explain experimental data (Hale, 2001).”
Incrementality and Prediction in Human Sentence Processing (2009, background): “…nor are our observations particularly unique; there is a growing body of computational work under the banner of expectation-based parsing, drawing on insights from information theory, which is predicated on parsing as a probabilistic and inherently predictive process (e.g., Hale, 2001; Levy, 2008).”
Redundancy and reduction: Speakers manage syntactic information density (2010, background): “This links the results reported here to findings suggesting that syntactic comprehension is probability-sensitive (Hale, 2001; Hale, 2003; Jurafsky, 1996; Kamide et al., 2003; Levy, 2008; MacDonald, 1994; McDonald & Shillcock, 2003; Staub & Clifton, 2006; Trueswell, 1996).”
Consequences of the Now-or-Never bottleneck for signed versus spoken languages (2015, background): “…interest in rational, rather processing-based, explanations in cognitive science (e.g., Anderson 1990; Chater et al. 2006 Griffiths & Tenenbaum 2009; Oaksford & Chater 1998; 2007; Tenenbaum et al. 2011), including language processing (Gibson et al. 2013; Hale 2001; 2006; Piantadosi et al. 2011).”
How do forward models work? And why would you want them? (2013, positive): “But although we certainly concur that prediction plays an important explanatory role for theories of language, we cannot help feeling that the emphasis given to action-based prediction in this model – and prediction in general throughout much of recent psycholinguistics (Altmann & Mirkovic ́ 2009; D”
What do we mean by prediction in language comprehension? (2015, background): “times (Hale, 2001; Levy, 2008) and (b) neural activity associated with processing them (Frank, Otten, Galli, & Vigliocco, 2015).”
The neural architecture of language: Integrative modeling converges on predictive processing (2020, methodology): “…models of sentence processing have previously been used to explain both behavioral (Dotlačil, 2018; Futrell, Gibson, & Levy, 2020; Gibson, 1998; Gibson et al., 2013; Hale, 2001; Jurafsky, 1996; Figure 1: Comparing Artificial Neural Network models of language processing to human language processing.”
What do RNN Language Models Learn about Filler–Gap Dependencies? (2018, background): “It is known to correlate directly with human sentence processing difﬁculty (Hale, 2001; Levy, 2008; Smith and Levy, 2013).”