Learning probabilities over underlying representations
نویسندگان
چکیده
We show that a class of cases that has been previously studied in terms of learning of abstract phonological underlying representations (URs) can be handled by a learner that chooses URs from a contextually conditioned distribution over observed surface representations. We implement such a learner in a Maximum Entropy version of Optimality Theory, in which UR learning is an instance of semisupervised learning. Our objective function incorporates a term aimed to ensure generalization, independently required for phonotactic learning in Optimality Theory, and does not have a bias for single URs for morphemes. This learner is successful on a test language provided by Tesar (2006) as a challenge for UR learning. We also provide successful results on learning of a toy case modeled on French vowel alternations, which have also been previously analyzed in terms of abstract URs. This case includes lexically conditioned variation, an aspect of the data that cannot be handled by abstract URs, showing that in this respect our approach is more general.
منابع مشابه
First Order Markov Decision Processes A Dissertation submitted
Relational Markov Decision Processes (RMDP) are a useful abstraction for complex reinforcement learning problems and stochastic planning problems since one can develop abstract solutions for them that are independent of domain size or instantiation. This thesis develops compact representations for RMDPs and exact solution methods for RMDPs using such representations. One of the core contributio...
متن کاملThe extraction and integration framework: a two-process account of statistical learning.
The term statistical learning in infancy research originally referred to sensitivity to transitional probabilities. Subsequent research has demonstrated that statistical learning contributes to infant development in a wide array of domains. The range of statistical learning phenomena necessitates a broader view of the processes underlying statistical learning. Learners are sensitive to a much w...
متن کاملSymmetry in Probabilistic Databases
Researchers in databases, AI, and machine learning, have all proposed representations of probability distributions over relational databases (possible worlds). In a tuple-independent probabilistic database, the possible worlds all have distinct probabilities, because the tuple probabilities are distinct. In AI and machine learning, however, one typically learns highly symmetric distributions, w...
متن کاملTransitional Probabilities Are Prioritized over Stimulus/Pattern Probabilities in Auditory Deviance Detection: Memory Basis for Predictive Sound Processing.
UNLABELLED Representations encoding the probabilities of auditory events do not directly support predictive processing. In contrast, information about the probability with which a given sound follows another (transitional probability) allows predictions of upcoming sounds. We tested whether behavioral and cortical auditory deviance detection (the latter indexed by the mismatch negativity event-...
متن کاملLearning Expected Hitting Time Distance
Most distance metric learning (DML) approaches focus on learning a Mahalanobis metric for measuring distances between examples. However, for particular feature representations, e.g., histogram features like BOW and SPM, Mahalanobis metric could not model the correlations between these features well. In this work, we define a nonMahalanobis distance for histogram features, via Expected Hitting T...
متن کامل