Error-driven Learning in Ot and Hg: a Comparison
نویسنده
چکیده
The OT error-driven learner is known to admit guarantees of efficiency, stochastic tolerance and noise robustness which hold independently of any substantive assumptions on the constraints. This paper shows that the HG learner instead does not admit such constraint-independent guarantees. The HG theory of error-driven learning thus needs to be substantially restricted to specific constraint sets.
منابع مشابه
Noise robustness and stochastic tolerance of OT error-driven ranking algorithms
Recent counterexamples show that Harmonic Grammar (HG) error-driven learning (with the classical Perceptron reweighing rule) is not robust to noise and does not tolerate the stochastic implementation (Magri 2014, MS). This article guarantees that no analogous counterexamples are possible for proper Optimality Theory (OT) error-driven learners. In fact, a simple extension of the OT convergence a...
متن کاملError-driven Learning in Harmonic Grammar
The HG literature has adopted so far the Perceptron reweighing rule because of its convergence guarantees. Yet, this rule is not suited to HG, as it fails at ensuring non-negativity of the weights. The first contribution of this paper is a solution to this impasse. I consider a variant of the Perceptron which truncates any update at zero, thus maintaining the weights non-negative in a principle...
متن کاملHG has no computational advantages over OT: consequences for the theory of OT online algorithms
Various authors have recently endorsed Harmonic Grammar (HG) as a replacement of Optimality Theory (OT). One argument for this move is based on computational considerations: OT looks prima facie like an exotic framework with no correspondent in Machine Learning, and the replacement with HG allows methods and results form Machine Learning to be imported within Computational Phonology; see for in...
متن کاملHarmonic Grammar, Optimality Theory, and Syntax Learnability: An Empirical Exploration of Czech Word Order
This work presents a systematic theoretical and empirical comparison of the major algorithms that have been proposed for learning Harmonic and Optimality Theory grammars (HG and OT, respectively). By comparing learning algorithms, we are also able to compare the closely related OT and HG frameworks themselves. Experimental results show that the additional expressivity of the HG framework over O...
متن کاملConvergence of Error-driven Ranking Algorithms
According to the OT error-driven ranking model of language acquisition, the learner performs a sequence of slight re-rankings triggered by mistakes on the incoming stream of data, until it converges to a ranking that makes no more mistakes. This learning model is very popular in the OT acquisition literature, in particular because it predicts a sequence of rankings that models gradualness in ch...
متن کامل