Comment Re:Confessions of a UW English major. (Score 1) 607
Whether or not a person can correctly edit a document is hardly a realistic measure of the quality of a software package.
For instance, how many people can solve partial differential equations? The vast majority of people will look at you with a dopey look in response to such a question. Writing software to solve such mathematical wonders is definitely possible though, with 100% correctness (software bugs withstanding - the point is that the intended behaviour is simply modelled incorrectly by the developer, not the intended behaviour representing something that cannot be modelled perfectly).
Whether or not someone needs to look up grammatical terms for clarification is irrelevant when dealing with software, as computers tend to be quite good at 'looking things up'.
Models able to tout 100% accuracy for classification simply do not exist. Computational linguists have been working on this for a long time, and they've not produced a solution yet.
For models that do approach 90% accuracy and better, the sheer complexity (both space and time) is too high for consumer applications. Requiring hours of computing time to initialise a model from training data is normal. To save the internal state of the model so that the training stage can be avoided requires massive amounts of storage and thus a gnarly loading time to contend with.
For instance, how many people can solve partial differential equations? The vast majority of people will look at you with a dopey look in response to such a question. Writing software to solve such mathematical wonders is definitely possible though, with 100% correctness (software bugs withstanding - the point is that the intended behaviour is simply modelled incorrectly by the developer, not the intended behaviour representing something that cannot be modelled perfectly).
Whether or not someone needs to look up grammatical terms for clarification is irrelevant when dealing with software, as computers tend to be quite good at 'looking things up'.
Models able to tout 100% accuracy for classification simply do not exist. Computational linguists have been working on this for a long time, and they've not produced a solution yet.
For models that do approach 90% accuracy and better, the sheer complexity (both space and time) is too high for consumer applications. Requiring hours of computing time to initialise a model from training data is normal. To save the internal state of the model so that the training stage can be avoided requires massive amounts of storage and thus a gnarly loading time to contend with.
