S. Kirby and J.R. Hurford. The evolution of incremental learning: language, development and critical periods. Edinburgh Occasional Papers in Linguistics, September 22 1997.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....which first uses simplified data, would allow the network to learn an inappropriate data representation. The network would later experience difficulties in adjusting to more complex data. Similarly Zeng [46] indicated to have observed a negative influence of incremental learning. Kirby Hurford [23] associate the term incremental learning with . the idea that some learning related resource starting at a low value, which then gradually increases while (but not necessarily because) the organism matures . and . the initial low (immature) value of resource actually ....

....necessarily because) the organism matures . and . the initial low (immature) value of resource actually facilitates, or even enables, the early stages of learning. Later stages of learning are in turn facilitated, or enabled, by higher valued settings of the resource concerned . [23] They note that the elements of timing and evolution are missing from Elman s models [8] Kirby Hurford [23] further claim that . any innate pre programming for input sensitive growth must have evolved . and they propose a method to incorporate the less is more [29] and starting ....

[Article contains additional citation context not shown here]

S. Kirby and J.R. Hurford. The evolution of incremental learning: language, development and critical periods. Edinburgh Occasional Papers in Linguistics, September 22 1997.

.... have included (proto)language(s) capable of creating (natural) selection pressure in favour of more efficient and reliable language acquisition (e.g. Kirby, 1998) In turn, the evolving language acquisition device created (linguistic) selection pressure in favour of more learnable languages (e.g. Kirby and Hurford, 1997a) Inductive bias in language acquisition emerges as a direct consequence of this coevolutionary process. Section 1.1 briefly surveys the current state of research on grammatical acquisition and provides the theoretical background which motivates the acquisition model presented in section 2. ....

.... model builds in no strong assumptions about the function(s) of language, whether this be to influence others, communicate (mis)information, or whatever (see e.g. Pinker and Bloom, 1990; Keller, 1994:84f for insightful discussion) 25 The critical period is simply stipulated Hurford (1991) and Kirby and Hurford (1997b) describe simulations in which such a criticial period emerges given certain assumptions. 23 1. Generate cost: 1 (GC) 2. Generate subset language cost: GSC(g) 3. Parse cost: 1 (PC) 4. Parse failure cost: 1 (PF) 5. Parse memory cost: WML(stjg) 6. Parse Generate success benefit: 1 (SI) 7. ....

[Article contains additional citation context not shown here]

Kirby, Simon and Jim Hurford (1997) The evolution of incremental learning: language, development and critical periods, Edinburgh Occasional Papers in Linguistics, 97-2.

.... relevant parameters (see above) A series of sim 14 A modification of this type might also form the basis of a less stipulative version of the critical period for learning in which LAgts simply ceased to track posterior probabilities of parameters once they reached threshold; see, e.g. Kirby and Hurford, 1997 for discussion and putative explanations of the critical period for language acquisition. 15 Languages which do not have attested counterparts are given mnemonic names in which F means the absence of parameter F and F indicates a marked parameter value with respect to the main attested full ....

Hurford, J. and Kirby, S. (1997) The evolution of incremental learning: language, development and critical periods, Edinburgh Occasional Papers in Linguistics, 97-2.

.... subset language speakers emerge, the SOVv2 LAgts 8 A modification of this type might also form the basis of a less stipulative version of the critical period for learning in which LAgts simply ceased to track posterior probabilities of parameters once they reached threshold; see, e.g. Kirby and Hurford, 1997 for discussion and putative explanations of the critical period for language acquisition. 9 Languages which do not have attested counterparts are given mnemonic names in which F means the absence of parameter F and F indicates a marked parameter value with respect to the main attested full ....

Hurford, J. and Kirby, S. (1997) The evolution of incremental learning: language, development and critical periods, Edinburgh Occasional Papers in Linguistics, 97-2.

....explore the space defined by this initial variation. And 12 A modification of this type might also form the basis of a less stipulative version of the critical period for learning in which LAgts simply ceased to track posterior probabilities of parameters once they reached threshold; see, e.g. Hurford and Kirby, 1997 for discussion and putative explanations of the critical period for language acquisition. secondly, by also using a mutation operator during reproduction which can introduce variation during a simulation run, with reproduction via crossover propagating successful mutations through the ....

Hurford, J. and Kirby, S. (1997) The evolution of incremental learning: language, development and critical periods, Edinburgh Occasional Papers in Linguistics, 97-2.

....and optionally mutation of the n updatable parameters per trigger. Crossover and mutation can turn an absolute (inherited) principle into a default or unset parameter and vice versa, change values of either, and so forth. 21 19 The critical period is simply stipulated Hurford (1991) and Hurford and Kirby (1997) describe evolutionary simulations in which such a criticial period emerges given certain assumptions. 20 The encoding of p settings allows the deterministic recovery of the initial setting because reset parameters are those preceded by R , or followed by a determinate value. parameters ....

Hurford, J. and Kirby, S. (1997) The evolution of incremental learning: language, development and critical periods, Edinburgh Occasional Papers in Linguistics, 97-2.

.... also communicate less successfully than SOVv2 13 A modification of this type might also form the basis of a less stipulative version of the critical period for learning in which LAgts simply ceased to track posterior probabilities of parameters once they reached threshold; see, e.g. Kirby and Hurford, 1997 for discussion and putative explanations of the critical period for language acquisition. 14 Languages which do not have attested counterparts are given mnemonic names in which F means the absence of parameter F and F indicates a marked parameter value with respect to the main attested full ....

Hurford, J. and Kirby, S. (1997) The evolution of incremental learning: language, development and critical periods, Edinburgh Occasional Papers in Linguistics, 97-2.

....can be analysed with the new grammar, adopt the new grammar. 5. Otherwise keep the current grammar. 4.6 Natural selection In order to implement natural selection we need some way of assessing the communicative success of individuals after learning. We use the concept of a critical period [27, 26] during which learning occurs, which is followed by a period of continued language use, but no grammatical change. As a simplifying assumption we measure the communicative fitness of individuals after this critical period. The fitness can be based on either the transmission ability of an ....

....the risk of failing to converge on the correct grammar before the critical period. This is why, when the critical period is changed, the degree of eventual nativisation changed. Only when the critical period is extremely severe do we see a complete nativisation of the functional constraints. See [26] for a model of how the critical period itself evolves. In summary, we have a two stage process: 1. From initially random initial conditions, linguistic selection leads to a glossogenetic adaptation of the languages in the arena of use. This results in observable constraints on variation, ....

Simon Kirby and James Hurford. The evolution of incremental learning: Language, development and critical periods. Occasional Paper EOPL-97-2, Department of Linguistics, University of Edinburgh, Edinburgh, 1997.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC