Predicting is not explaining: targeted learning of the dative alternation - HAL-SHS - Sciences de l'Homme et de la Société Accéder directement au contenu
Article Dans Une Revue Journal of Causal Inference Année : 2016

Predicting is not explaining: targeted learning of the dative alternation

Antoine Chambaz
Guillaume Desagulier

Résumé

Corpus linguists dig into large-scale collections of texts to better understand the rules governing a given language. We advocate for ambitious corpus linguistics drawing inspiration from the latest developments of semiparametrics for a modern targeted learning. Transgressing discipline-specific borders, we adapt an approach that has proven successful in biostatistics and apply it to the well-travelled case study of the dative alternation in English. A dative alternation is characterized by sentence pairs with the same verb, but different syntactic patterns, e.g. I gave a book to him (prepositional dative) and I gave him a book (double-object dative). Our aim is to explain how native speakers of English choose a pattern over another in any given context. The essence of the approach hinges on causal analysis and targeted minimum loss estimation (TMLE). Through causal analysis, we operationalize the set of scientific questions that we wish to address regarding the dative alternation. Drawing on the philosophy of TMLE, we answer these questions by targeting some versatile machine learners. We derive estimates and confidence regions for well-defined parameters that can be interpreted as the influence of each contextual variable on the outcome of the alternation (prepositional vs. double-object), all other things being equal.
Fichier principal
Vignette du fichier
chambazDesagulier_dative_rev1.pdf (418.77 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)
Loading...

Dates et versions

hal-01073005 , version 1 (08-10-2014)
hal-01073005 , version 2 (25-03-2015)

Identifiants

  • HAL Id : hal-01073005 , version 2

Citer

Antoine Chambaz, Guillaume Desagulier. Predicting is not explaining: targeted learning of the dative alternation. Journal of Causal Inference, 2016, 4 (1), pp.1-30. ⟨hal-01073005v2⟩
337 Consultations
471 Téléchargements

Partager

Gmail Facebook X LinkedIn More