Page Karen M
Department of Computer Science, Bioinformatics Unit, University College London, Gower Street, WC1E 6BT London, UK.
Bull Math Biol. 2004 Jul;66(4):651-62. doi: 10.1016/j.bulm.2003.09.007.
In order to learn grammar from a finite amount of evidence, children must begin with in-built expectations of what is grammatical. They clearly are not born, however, with fully developed grammars. Thus early language development involves refinement of the grammar hypothesis until a target grammar is learnt. Here we address the question of how much evidence is required for this refinement process, by considering two standard learning algorithms and a third algorithm which is presumably as efficient as a child for some value of its memory capacity. We reformulate this algorithm in the context of Chomsky's 'principles and parameters' and show that it is possible to bound the amount of evidence required to almost certainly speak almost grammatically.
为了从有限的证据中学习语法，儿童必须从对什么是符合语法的内在期望开始。然而，他们显然并非生来就具备完全成熟的语法。因此，早期语言发展涉及对语法假设的细化，直到学会目标语法。在这里，我们通过考虑两种标准学习算法和第三种算法（对于其记忆容量的某些值，该算法可能与儿童一样高效）来解决这个细化过程需要多少证据的问题。我们在乔姆斯基的“原则与参数”框架下重新表述了这个算法，并表明几乎可以肯定地说出几乎符合语法的话语所需的证据量是可以界定的。
