6 Distributed computational models of intervention effects: A study on cleft structures in French
-
Abstract
Object it-cleft constructions are complex structures which also occur rarely in corpora. When present, it has been demonstrated that there is a crosslinguistic tendency to disfavour matching configurations in terms of intervention effects triggered by morpho-syntactic features between the fronted clefted element and the intervening subject. If the investigation in large scale-corpora suggest that similarity between the clefted fronted object and the intervening subject is avoided, we expect that computational models sensitive to the statistics might show a dispreference for matching and a preference for mismatching configurations as predicted from a theory of locality. In this paper, we carry out two studies on artificial neural network models, which represent powerful domain-general learning mechanisms with weak learning biases, trained in French. What we observe is that the representations of neural network models are sensitive to morpho-syntactic features (type/number/person and number/gender), since we observe gradation of effects that vary with the number of matching features.
Abstract
Object it-cleft constructions are complex structures which also occur rarely in corpora. When present, it has been demonstrated that there is a crosslinguistic tendency to disfavour matching configurations in terms of intervention effects triggered by morpho-syntactic features between the fronted clefted element and the intervening subject. If the investigation in large scale-corpora suggest that similarity between the clefted fronted object and the intervening subject is avoided, we expect that computational models sensitive to the statistics might show a dispreference for matching and a preference for mismatching configurations as predicted from a theory of locality. In this paper, we carry out two studies on artificial neural network models, which represent powerful domain-general learning mechanisms with weak learning biases, trained in French. What we observe is that the representations of neural network models are sensitive to morpho-syntactic features (type/number/person and number/gender), since we observe gradation of effects that vary with the number of matching features.
Chapters in this book
- Frontmatter I
- Contents V
- It-clefts: State-of-the-art, and some empirical challenges 1
- 1 Cleft wh-questions as biclausal structures 11
- 2 What is it that requires or constrains clefts? (Dis)Favouring factors for clefting in Germanic and Romance 35
- 3 Subject versus object clefts: A fresh perspective on a robust asymmetry 81
- 4 Making the case for distinguishing information structure from specification in English it-clefts 105
- 5 The emergence and early development of c’est ‘it is’ clefts in French L1 135
- 6 Distributed computational models of intervention effects: A study on cleft structures in French 157
- 7 It-cleft constructions in Réunion Creole 181
- 8 (It-)clefts in Palenquero Creole and the specificational copula 217
- 9 A cartographic approach to Chinese V de O clefts 235
- Index 257
Chapters in this book
- Frontmatter I
- Contents V
- It-clefts: State-of-the-art, and some empirical challenges 1
- 1 Cleft wh-questions as biclausal structures 11
- 2 What is it that requires or constrains clefts? (Dis)Favouring factors for clefting in Germanic and Romance 35
- 3 Subject versus object clefts: A fresh perspective on a robust asymmetry 81
- 4 Making the case for distinguishing information structure from specification in English it-clefts 105
- 5 The emergence and early development of c’est ‘it is’ clefts in French L1 135
- 6 Distributed computational models of intervention effects: A study on cleft structures in French 157
- 7 It-cleft constructions in Réunion Creole 181
- 8 (It-)clefts in Palenquero Creole and the specificational copula 217
- 9 A cartographic approach to Chinese V de O clefts 235
- Index 257
