add lessons with linguistic objects

[jeisner]

Right now the lessons are not very NLP-ish. We had originally talked about having some lessons that illustrate maxent models in NLP, on tiny examples.
Below is a quote from my email of 9/10/12 when we were doing initial planning.
I've added the material in [brackets].

---

ability to show text [and resizable images] rather than shapes? Would be splendid to illustrate some NLP applications of conditional modeling:

• predict sentiment (+, -) or topic conditioned on text; here the shapes can be big "+" and "-" and the features can conjoin [we've already done a version of this, but could probably improve by using sentiment rather than spam, with smiley and frowny shapes. Should also improve explanation of how binary logistic regression can get by with only the + features, but multi-class has to conjoin with each possible class.]
• predict direct object given verb (for tiny vocab) - this is a simplified version of language modeling
• predict syntax tree given string [there would be little images of 4 trees for a very short sentence, and the images would get bigger and smaller as one manipulated the rule weights. Possibly each tree might appear inside a circle or square whose area was being controlled.]
• [distribution over syntactic rules given LHS, using things like plural agreement]
• [WSD using contextual features: each sense would be indicated by a paraphrase word or a picture of an object, e.g., a riverbank vs. a financial bank]
• predict correct spelling given incorrect spelling and context
• [predict segmentation given an ambiguous string in English or Chinese -- the features would be word unigrams]
• [predict tagging given a string]

to support the examples above: for each object in each context, perhaps file format should give its visual appearance and separately list the names of the features that it has. Then create a slider for each feature that appears at least once in the dataset.

[fmof]

The current solution to #64 should allow progress on this.

[fmof]

See #69.