Language comprehension as a multi-label classification problem

Konstantin Sering, Petar Milin, R. Harald Baayen

Research output: Contribution to journalArticlepeer-review

4 Citations (Scopus)


The initial stage of language comprehension is a multilabel classification problem. Listeners or readers, presented with an utterance, need to discriminate between the intended words and the tens of thousands of other words they know. We propose to address this problem by pairing two net- works. The first network is independently learned with the Rescorla Wagner model. The second network is based on the first network and learned with the rule of Widrow and Hoff. The first network has to recover from sublexical input features the meanings encoded in the language signal, resulting in a vector of activations over the lexicon. The sec- ond network takes this vector as input and further reduces uncertainty about the intended message. Classification per- formance for a lexicon with 52,000 entries is good. The model also correctly predicts several aspects of human lan- guage comprehension. By rejecting the traditional linguis- tic assumption that language is a (de)compositional sys- tem, and by instead espousing a discriminative approach, a more parsimonious yet highly effective functional charac- terization of the initial stage of language comprehension is obtained.
Original languageEnglish
Article number72
Pages (from-to)339-353
JournalStatistica Neerlandica
Early online date19 Apr 2018
Publication statusPublished - 1 Aug 2018


  • error-driven learning
  • language comprehension
  • multilabel classification
  • Rescorla-Wagner
  • Widrow-Hoff


Dive into the research topics of 'Language comprehension as a multi-label classification problem'. Together they form a unique fingerprint.

Cite this