scholarly journals A noisy-channel model of rational human sentence comprehension under uncertain input

Author(s):  
Roger Levy
2021 ◽  
Author(s):  
Nina Zdorova ◽  
Svetlana Malyutina ◽  
Anna Laurinavichyute ◽  
Anastasiia Kaprielova ◽  
Kromina Anastasia ◽  
...  

Noise, as part of real-life communication flow, degrades the quality of linguistic input and affects language processing. According to predictions of the noisy-channel model, noisemakes comprehenders rely more on word-level semantics and good-enough processing instead of actual syntactic relations. However, empirical evidence of such qualitative effect of noise on sentence processing is still lacking. For the first time, we investigated the qualitative effect of both auditory (three-talker babble) and visual (short idioms appearing next to target sentence on the screen) noise on sentence reading within one study in two eye-trackingexperiments. In both of them, we used the same stimuli — unambiguous grammatical Russian sentences — and manipulated their semantic plausibility. Our findings suggest that although readers relied on good-enough processing in Russian, neither auditory nor visualnoise qualitatively increased reliance on semantics in sentence comprehension. The only effect of noise was found in reading speed: only without noise, semantically implausible sentences were read slower than semantically plausible ones, as measured by both early and late eye-movement measures. These results do not support the predictions of the noisy-channel model. With regard to quantitative effects, we found a detrimental effect ofauditory noise on overall comprehension accuracy, and an accelerating effect of visual noise on sentence processing without accuracy decrease.


2013 ◽  
Vol 61 (1) ◽  
pp. 390-402 ◽  
Author(s):  
Ai-ichiro Sasaki ◽  
Takako Ishihara ◽  
Nobutaro Shibata ◽  
Ryusuke Kawano ◽  
Hiroki Morimura ◽  
...  

2012 ◽  
Vol 20 (6) ◽  
pp. 1784-1794 ◽  
Author(s):  
Daisuke Saito ◽  
Shinji Watanabe ◽  
Atsushi Nakamura ◽  
Nobuaki Minematsu

2017 ◽  
Author(s):  
Melissa Kline ◽  
Miguel Angel Salinas ◽  
Eunice Lim ◽  
Evelina Fedorenko ◽  
Edward Gibson

A fundamental typological variation in the world’s languages is their basic word order; around 80% of spoken languages are either Subject-Object-Verb (SOV) or Subject-Verb-Object (SVO). Previous work has related this typological pattern to a striking finding in ad-hoc gesture production: across a wide range of languages, people tend to use the order SOV to gesture events with inanimate patients, and SVO orders for those with animate patients (Gibson, Piantadosi, et al., 2013; Hall, Mayberry, & Ferreira, 2013, i.a.). Gibson et al. (2013) interpret this evidence as support for a noisy channel model of communication, under which producers attempt to reduce ambiguity for comprehenders. On the other hand, this pattern might also result from the particular kinds of gestures that people tend to use for different kinds of events (Hall et al. 2013). We conducted two production studies – one in gesture, one using non-iconic written symbols– designed to modify the communication task in different ways. Two main findings emerged from these tests: first, simple modifications to the gesture paradigm can have a profound effect on the orders used: the instructions given in Experiment 1 (gesture) dramatically affected the use of SVO orders, even though participants were gesturing about the same events under the same communicative contexts. Second, there is an apparent association between the physical form of the gestures used to communicate and word order: we found that participants were most likely to use SVO orders in Experiment 1 for animate-animate events when using body-based gestures (avoiding role conflict, as described by Hall et al. 2013); furthermore in Experiment 2 (written symbols), there was no association between SVO ordering and animate-animate scenes (contrary to predictions of the noisy channel model). We conclude that gesturing paradigms, while a striking and naturalistic example of ad-hoc communication dynamics, are also sensitive to the particular modality and task instructions, and as such do not provide straightforward evidence for noisy channel theories of word order typology.


2014 ◽  
Author(s):  
Hsun-wen Chiu ◽  
Jian-cheng Wu ◽  
Jason S. Chang
Keyword(s):  

2013 ◽  
Vol 1 ◽  
pp. 125-138 ◽  
Author(s):  
Sam Sahakian ◽  
Benjamin Snyder

During the course of first language acquisition, children produce linguistic forms that do not conform to adult grammar. In this paper, we introduce a data set and approach for systematically modeling this child-adult grammar divergence. Our corpus consists of child sentences with corrected adult forms. We bridge the gap between these forms with a discriminatively reranked noisy channel model that translates child sentences into equivalent adult utterances. Our method outperforms MT and ESL baselines, reducing child error by 20%. Our model allows us to chart specific aspects of grammar development in longitudinal studies of children, and investigate the hypothesis that children share a common developmental path in language acquisition.


2021 ◽  
Vol 9 ◽  
pp. 657-674
Author(s):  
Qi Liu ◽  
Lei Yu ◽  
Laura Rimell ◽  
Phil Blunsom

Abstract Direct decoding for task-oriented dialogue is known to suffer from the explaining-away effect, manifested in models that prefer short and generic responses. Here we argue for the use of Bayes’ theorem to factorize the dialogue task into two models, the distribution of the context given the response, and the prior for the response itself. This approach, an instantiation of the noisy channel model, both mitigates the explaining-away effect and allows the principled incorporation of large pretrained models for the response prior. We present extensive experiments showing that a noisy channel model decodes better responses compared to direct decoding and that a two-stage pretraining strategy, employing both open-domain and task-oriented dialogue data, improves over randomly initialized models.


Sign in / Sign up

Export Citation Format

Share Document