Should I avoid adding sentences to vocab lists? Does it cause harm by adding garbage to some permanent global definition list intended only for words?

The problem about studying sentences is that they take ages to write out, and really slows down your study. Might be better to keep Skritter for words and characters, and put the sentences in a flashcards program like Anki.

But, if you want to go ahead an add one regardless, that's fine.

We're actually planning (eventually) to extend the system to allow studying sentences. As Byzanti points out, it's not efficient to write them out, and so we would probably have the parts you study for them limited, or change the way they are studied. We were thinking of limiting the parts for studying lists to readings (only flashcard style) and definitions. What do you guys think; how would Skritter teach you sentences, ideally? It's going to be a while before we build this but we can start talking about how it will work now.

But, for now, yeah probably should use Anki for sentences. We'd like to keep the dictionary to words and, at most, common phrases until we actually build this upcoming system.

Just doing flashcard-style reading and definitions for sentences is a good start; I would rather mark myself correct or incorrect rather than write out everything just to let Skritter's algorithm determine if I did things correctly or not. I also think sentences should be studied separately from words. It seems to me that the salient features of studying sentences are the learning of grammatical patterns and proper usage of specific words, so a tool for studying sentences would ideally isolate these things for study.

I'm going to start thinking aloud. Perhaps, to learn grammar patterns, it could be useful to have a template sentence with one or two variable words, such as an adjective, adverb, verb, or noun that is taken from the user's vocabulary and more or less randomly chosen. This way, when that pattern pops up for study, the words aren't always exactly the same but the grammar pattern is. Of course, this would be extremely difficult to implement: if select words are randomized in the sentence, you could end up studying nonsense like "I'm going to the panda bear tomorrow" rather than "I'm going to the zoo tomorrow," not to mention the fact that the algorithm for choosing words would need to take care of articles and tense agreement in English.

One of the biggest problems for me as a Chinese learner is trying to distinguish different words that seem synonymous. I think that seeing these words in context is the most efficient way to get a sense for how certain words are used. Perhaps, then, it could be useful to find articles or similar things from reliable sources, such as newspapers on the internet, and have an algorithm retrieve sentences which contain vocabulary that's already in My Words. This doesn't seem to be very straightforward either: you would need articles that also have sentence-by-sentence English translations, since machine translation isn't too stellar yet, and an algorithm to parse a sentence properly into words and check those words against My Words.

Okay, enough thinking aloud. I'm being a bit of a stubborn utopianist here by coming up with all these in-a-perfect-world ideas. Flashcards seem like a good start, and not too difficult compared to my ideas either.

We probably won't do the sort of madlibs approach of grammar patterns; that would be rather complex and probably not add too much benefit. Best to have concrete examples rather than randomly generated ones.

Actually, for sentence sources, we're planning on using Tatoeba. It's a large repository for example sentences between all different languages, and would be a great way to seed the system. We're still figuring out how exactly we would parse and organize that data the best way, but ideally you would be able to see all the sentences a given word is in and easily add some of them to your studies, just as you might add words with a given character in it now, from the word popup.

Is Tatoeba the source for the current example sentences system?

How feasible would it be to filter the tatoeba sentences to reject sentences that include more than one or two characters that I have not studied yet? (either as part of skritter or something that could be done outside of skritter, by, say, exporting my list of characters and doing some technical wizardry with the tatoeba data). That's been one of my biggest obstacles to sentence study, at least with the example sentences skritter uses now: If I want to study the sentence for the target word, I have to learn 4 or 5 other characters/words just to understand the example sentence.

No; the current sentences came from an effort from dict.cn.

That would be pretty tough. The current plan is to just pick the most popular sentences by default, and ordering by easiest sentence first when there are not enough votes yet. With those sentences, the easiest sentences should be much easier in general than the ones we have now.

I totally agree with GrandPoohBlah about sentences that should be studied separately from words and characters, because they are made of "most likely already known characters", and it's boring having to rewrite them entirely while studying definition would be enough.

Another reason is the graphical interface that is not suited (width), switching to a "flashcard mode" with sentences lists would fix the problem I think.

Yet another argument for studying sentences is that skritter definitions are minimalistics (for practical reasons), and it helps grasping a more accurate definition by providing context and usage examples; note that MDBG provides access to related sentences for each definition.

About grammar pattern, I think associating them with a set of well selected sentences examples + a definition would be more appropriate than treating them as a simple word, simply because they contain symbols (e.g. ...(dots)) which implies underlying grammar rules.