I've tried both ways in the past, all sentences, all single word, but to clarify some stuff I mentioned before, now I do 'both', but like so, with subs2srs decks: The 'sentence' cards are all in one deck with video clips. I don't target/prioritize vocabulary for those, I prioritize the audio and the grammar and understanding as a whole. So to prepare to do a set of new sentence cards in that deck, I take all the vocabulary I don't know from those sentences, and do those as single word cards in a separate deck--sometimes with audio from Breen's dictionary or TTS, maybe with pictures (haven't bothered with that in a while, though, too much trouble) I dragged from a Google image search.
That goes pretty quickly, and after I've passed them a couple times i.e. when they're somewhat mature, then I go back to the other deck and it's much easier to just work on parsing the grammar and focusing on listening, etc., for the entire sentence. In that sense, I have no problem picking up vocabulary quickly but still get some SRSed contextual benefits. It's just a matter of reducing overhead but trying to keep the complementary relationships.
To highlight some terms I think in: From: http://www-writing.berkeley.edu/tesl-ej/ej32/a1.htm - "Corpora come in many shapes and sizes, because they are built to serve different purposes. [3] There are two philosophies behind their design, leading to the distinction between reference and monitor corpora. Reference corpora have a fixed size; that is, they are not expandable (e.g., the British National Corpus), whereas monitor corpora are expandable; that is, texts are continuously being added (e.g., the Bank of English). Another design-related distinction is whether a corpus contains whole texts, or merely samples of a specified length. The latter option allows a greater variety of texts to be included in a corpus of a given size.
In terms of content, corpora can be either general, that is, attempt to reflect a specific language or variety in all its contexts of use (e.g., the American National Corpus), or specialised, that is, aim to focus on specific contexts and users (e.g., Michigan Corpus of Academic Spoken English), and they can contain written or spoken language. Corpora can also represent the different varieties of a single language. For example, the International Corpus of English (ICE) contains one-million-word corpora representative of different varieties of English (British, Indian, Singaporean, etc.). As implied in the previous section, corpora may contain language produced by native or non-native speakers (usually learners). Finally, corpora can be monolingual (i.e., contain samples of only one language), or multilingual. Multilingual corpora are of two types: they can contain the same text-types in different languages, or they can contain the same texts translated into different languages, in which case they are also known as parallel corpora (Hunston, 2002; Kennedy, 1998; McEnery & Wilson, 2001; Meyer, 2002)."
So at the moment, in essence, I've always got a vocabulary deck and a sentence deck. I think of them as a general/reference corpus (sentences) and a specialized/monitor corpus (words). Well actually my KO2001/smart.fm deck is more of a reference corpus as it doesn't 'grow' and I use it for kind of all-purpose stuff, and then you could think of all subs2srs stuff as specialized/monitor work with specific goals, that're always changing in size.
Then there's my extemporaneous stuff, where I just stick interesting words or small phrases into the vocabulary deck as I come across them.
Point being, it's working extremely well for me, but I think it works best post-foundation stage. Personally I prefer having those complementary aspects in the SRS, but I could see mass vocab working well alongside mass media input.
That goes pretty quickly, and after I've passed them a couple times i.e. when they're somewhat mature, then I go back to the other deck and it's much easier to just work on parsing the grammar and focusing on listening, etc., for the entire sentence. In that sense, I have no problem picking up vocabulary quickly but still get some SRSed contextual benefits. It's just a matter of reducing overhead but trying to keep the complementary relationships.
To highlight some terms I think in: From: http://www-writing.berkeley.edu/tesl-ej/ej32/a1.htm - "Corpora come in many shapes and sizes, because they are built to serve different purposes. [3] There are two philosophies behind their design, leading to the distinction between reference and monitor corpora. Reference corpora have a fixed size; that is, they are not expandable (e.g., the British National Corpus), whereas monitor corpora are expandable; that is, texts are continuously being added (e.g., the Bank of English). Another design-related distinction is whether a corpus contains whole texts, or merely samples of a specified length. The latter option allows a greater variety of texts to be included in a corpus of a given size.
In terms of content, corpora can be either general, that is, attempt to reflect a specific language or variety in all its contexts of use (e.g., the American National Corpus), or specialised, that is, aim to focus on specific contexts and users (e.g., Michigan Corpus of Academic Spoken English), and they can contain written or spoken language. Corpora can also represent the different varieties of a single language. For example, the International Corpus of English (ICE) contains one-million-word corpora representative of different varieties of English (British, Indian, Singaporean, etc.). As implied in the previous section, corpora may contain language produced by native or non-native speakers (usually learners). Finally, corpora can be monolingual (i.e., contain samples of only one language), or multilingual. Multilingual corpora are of two types: they can contain the same text-types in different languages, or they can contain the same texts translated into different languages, in which case they are also known as parallel corpora (Hunston, 2002; Kennedy, 1998; McEnery & Wilson, 2001; Meyer, 2002)."
So at the moment, in essence, I've always got a vocabulary deck and a sentence deck. I think of them as a general/reference corpus (sentences) and a specialized/monitor corpus (words). Well actually my KO2001/smart.fm deck is more of a reference corpus as it doesn't 'grow' and I use it for kind of all-purpose stuff, and then you could think of all subs2srs stuff as specialized/monitor work with specific goals, that're always changing in size.
Then there's my extemporaneous stuff, where I just stick interesting words or small phrases into the vocabulary deck as I come across them.
Point being, it's working extremely well for me, but I think it works best post-foundation stage. Personally I prefer having those complementary aspects in the SRS, but I could see mass vocab working well alongside mass media input.
Edited: 2010-02-11, 5:57 pm