Probabilistic language models in cognitive neuroscience: promises and pitfalls

Armeni, Kristijan; Willems, Roel M.; Frank, Stefan

doi:10.1101/168161

Cited by 21 publications

(24 citation statements)

References 96 publications

(118 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Uncertainty and probabilistic computation have received increased attention as viable paradigms for describing information processing at the cognitive (Chater,740 Tenenbaum, & Yuille, 2006) and neural levels (Hasson, 2017;Knill & Pouget, 2004 processing (see Armeni, Willems, & Frank, 2017, for discussion).…”

Section: Discussionmentioning

confidence: 99%

“…Presently, the work employing these metrics (present study included) is not addressing the question of neural codes, that is, how probabilistic knowledge and functions needed for 750 language understanding can be encoded and decoded with models of neural circuits. They serve, instead, in describing the basic phenomenon-a statistical relationship between cognitive variables and neurophysiological observableswhich in itself requires explanation (see Armeni et al, 2017;Carlson, Goddard, Kaplan, Klein, & Ritchie, 2017;Shagrir & Bechtel, 2017, for similar remarks).…”

Section: Discussionmentioning

confidence: 99%

See 1 more Smart Citation

Frequency-specific brain dynamics related to prediction during language comprehension

Armeni

Willems

Bosch

et al. 2018

Preprint

Self Cite

View full text Add to dashboard Cite

The brain's remarkable capacity to process spoken language virtually in real time requires fast and efficient information processing machinery.In this study, we investigated how frequency-specific brain dynamics relate to models of probabilistic language prediction during auditory narrative comprehension. We recorded MEG activity while participants were listening to auditory stories in Dutch. Using trigram statistical language models, we estimated for every word in a story its conditional probability of occurrence. On the basis of word probabilities, we computed how unexpected the current word is given its context (word perplexity) and how (un)predictable the current linguistic context is (word entropy). We then evaluated whether source-reconstructed MEG oscillations at different frequency bands are modulated as a function of these language processing metrics. We show that theta-band source dynamics are increased in high relative to low entropy states, likely reflecting lexical computations. * k.armeni@donders.ru.nl Beta-band dynamics are increased in situations of low word entropy and perplexity possibly reflecting maintenance of ongoing cognitive context. These findings lend support to the idea that the brain engages in the active generation and evaluation of predicted language based on the statistical properties of the input signal.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Discussionmentioning

confidence: 99%

Frequency-specific brain dynamics related to prediction during language comprehension

Armeni

Willems

Bosch

et al. 2018

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…The value of surprisal (S) indicates how unexpected a given word is on the bases of the preceding words 1 . In order to calculate the surprisal value associated to each word of the sentence, we used the algorithms developed by Roark 2 with a model of Probabilistic Context Free Grammar , where P ( w i ) corresponds to the probability of occurrence of the target wordand P ( w 1,2,…, i −1 ) to the probability of occurrence of the preceding words.…”

Section: Methodsmentioning

confidence: 99%

Electrophysiological correlates of syntactic structures

Artoni

d’Orio

Catricalà

et al. 2019

Preprint

View full text Add to dashboard Cite

Syntax is traditionally defined as a specifically human way to pair sound with meaning: words are assembled in a recursive way generating a potentially infinite set of sentences 1,2 . There can be different phrasal structures depending on the types of words involved, for example, "noun phrases" (NP), combining an article and a noun, vs. "verb phrases" (VP), combining a verb and a complement. Although it is known that the combination of an increasing number of words in sequences correlates with an increasing electrophysiological activity 3,4 , the specific electrophysiological correlates of the syntactic operation generating NPs vs. VPs remain unknown. A major confounding factor is the fact that syntactic information is inevitably intertwined with the acoustic information contained in words even during inner speech 5 . Here, we addressed this issue in a novel way by designing a paradigm to factor out acoustic information and isolate the syntactic component. In particular, we construed phrases that have exactly the same acoustic content but that are interpreted as NPs or VPs depending on their syntactic context (homophonous phrases). By performing stereo-electro-encephalographic (SEEG) recordings in epileptic patients 6 we show that VPs are associated with a higher activity in the high gamma band (150-300Hz frequency), an index of cortical activity associated with linguistic processing, with respect to NPs in multiple cortical areas in both hemispheres, including language areas and their homologous in the nondominant hemisphere. Our findings pave the way to a deeper understanding of the electrophysiological mechanisms underlying syntax and contribute to the ultimate far reaching goal of a complete neural decoding of linguistic structures from the brain 2 .Factoring sound out of a linguistic expression is a major empirical problem, since sound is inevitably intertwined with syntactic information 3,7 even during inner speech 5 . By crucially relying on Italian grammar we designed a novel protocol to circumvent this problem and measure the specific electrophysiological correlates of two basic syntactic structures. The stimuli were pairs of different sentences containing strings of two words with exactly the same acoustic information but completely different syntax (homophonous strings). More specifically, each pair contained an NP, resulting from syntactic combination of two lexical elements (a definite article and a noun), and a VP, resulting from the syntactic combination of two different types of lexical elements (a verb and a pronominal complement): the NP and the VP were pronounced in exactly the same way. In addition, each VP included a further crucial difference: the object of the verb, realized as a pronoun, was moved from its canonical position on the right of the verb to the left of the verb, a syntactic operation called "cliticization". For example, the sequence [laˈpɔrta], could be interpreted either as a noun phrase ("the door") or a verb phrase ("brings her"; lit.: her brings) depending on the syntactic ...

show abstract

“…The value of surprisal (S) indicates how unexpected a given word is on the bases of the preceding word 27 . In order to calculate the surprisal value associated to each word of the sentence, we used the algorithms developed to the probability of occurrence of the preceding words.…”

Section: Surprisal Value Computationmentioning

confidence: 99%

Disentangling sound from syntax: electrophysiological analysis of linguistics expressions

Artoni¹,

d’Orio²,

Catricalà

et al. 2019

Preprint

View full text Add to dashboard Cite

Syntax is a species-specific component of human language combining a finite set of words in a potentially infinite number of sentences. Since words are by definition expressed by sound, factoring out syntactic information is normally impossible. Here, we circumvented this problem in a novel way by designing phrases with exactly the same acoustic content but different syntactic structures depending on the other words they occur with. By performing stereoelectroencephalographic (SEEG) recordings in epileptic patients we measured a different electrophysiological correlate of verb phrases vs. noun phrases by analyzing the high gamma band activity (150-300Hz frequency), in multiple cortical areas in both hemispheres, including language areas and their homologous in the non-dominant hemisphere. Our findings contribute to the ultimate goal of a complete neural decoding of linguistic structures from the brain.

show abstract

Probabilistic language models in cognitive neuroscience: promises and pitfalls

Cited by 21 publications

References 96 publications

Frequency-specific brain dynamics related to prediction during language comprehension

Frequency-specific brain dynamics related to prediction during language comprehension

Electrophysiological correlates of syntactic structures

Disentangling sound from syntax: electrophysiological analysis of linguistics expressions

Contact Info

Product

Resources

About