Talk:Weighted context-free grammar

Proposed merge with Stochastic context-free grammar
Weighted and stochastic CFGs are equivalent, as established by the WCFG article. We could discuss both in one article. Q VVERTYVS (hm?) 13:37, 11 February 2015 (UTC)

The WCFG article is incredibly short and contributes almost no additional information beyond what is in the stochastic context-free grammar page. A merge would be a good idea.

Note however that the name "stochastic context-free grammar" is not the standard one that is employed in computational linguistics. There, it is normal to refer to them as "probabilistic context-free grammars". This name is highly standardized; so standard, in fact, that it is actually the custom to refer to them with the acronym PCFG.

Also, it is weird that the stochastic context-free grammar article goes into such depth on the applications of PCFGs to protein matching, while barely mentioning their applications in computational linguistics and natural language processing. For example, the Penn Treebank is represented using PCFG-like representations, and this corpus has provided the training data for a healthy fraction of the NLP technologies in common commerical use. It is weird to me, to the point of bizarre, that neither context-free grammar page does not contain a single instance of a "tree" illustrating the intuitive graphical representation of a PCFG.

Last but not least, note that weighted CFGs are not actually equivalent to stochastic CFGs in all cases. A stochastic CFG always assigns a probability to a string; generally one is interested either in the maximum of the probabilities over all possible trees or the sum of the probabilities of all such trees. The probability must be between 0 and 1, and under relatively mild conditions you have that the sum of probabilities of all possible trees is equal to 1. In the more general case of weighted CFGs, it is possible for the maximum and/or sum to not be finite (although such cases are of limited practical use). Rtd885 (talk) 02:12, 1 March 2015 (UTC)
 * Agree with merge, and merging to separate section (given the distinction between them). Note also the target page name change that occurred in the interim (doesn't change the arguments for the merge). Klbrain (talk) 00:57, 17 September 2017 (UTC)