Introduction
As you might have guessed from the title, o+1 refers to comprehensible output. This is one of the four strands of language learning in the structure proposed by Paul Nation.
Comprehensible output or o+1, the output hypothesis, is the complement of i+1, or comprehensible input. The i or o refers to input or output. ‘Comprehensible’ in this case refers not to the comprehensibility of your output to others, but output that extends from your current level of comprehension. Output is a verb, here, a conversion process. Thus the +1 refers to this extension just beyond your current level. That is, the input or the output (i or o) should be +1 (just beyond your current level). Note that this doesn’t mean specifically that your output or your input should only contain 1 new vocabulary word. This is just an interpretation specific to studying digital flashcards that some learners use.
Years ago, I started using i+N, where N represents a variable number, to refer to the fact that the +1 is something abstract and changing as ‘chunking’ occurs. Somewhat amusingly I think this caused confusion, as i+1 is often mistakenly written now as N+1, which typically refers to other things in language-related work, such as perceptual span (where N is the fixated word and 1 is the word beyond it).
Like kanji handwriting practice but on a larger scale, output isn’t something that’s just for people who want to speak and write Japanese; it’s also for strengthening your receptive proficiency, including your listening and reading abilities, vocabulary, and grammar. Nor does it require interaction and discomfort to practice, as we shall see.
Learning through input alone is relatively slow and fragile. As Nation and other researchers have noted, words are learned through input at a very low rate, even with graded readers and glossing, and learning a word through context requires knowing around 98% of any given stream of input, and fully learning it this way requires seeing the word many times (10-20) in different contexts. As noted in the above links, incidental encounters have to be fairly frequent as their encoding is so impartial and imprecise, especially if the text is less than 98% comprehensible. You’d have to read 8 million words of text at your level just to learn 2000 words, at a rate of at least 1-2 graded readers a week. And then there’s grammar!
Thus input, while essential, is best used to flesh out deliberately learned aspects of the language (e.g. explicit grammar study, vocabulary words via SRS), as part of a balanced program. Just as monolingualism has caused an unscientific and impractical recommendation to avoid using the L1 (first language) and needs to be replaced with a multilingual perspective, overemphasis on input (due to theories based on outdated, pseudoscientific notions of a language organ in the brain, an acquisition/learning barrier, etc.) has caused many to avoid output, leaving critical weaknesses, one of which requires an o+1 approach to make right.
There’s also the marginalization of fluency practice, another aspect of Nation’s Four Strands. Fluency and fluency development refers to the fluid, spontaneous manner in which you can use the language items that you have learned so far. It doesn’t refer to 100% perfection of 100% of the language. It can refer to how fluidly and spontaneously you can use the first 10 words you’ve learned. Rather than the nonsensical monolingualist terms ‘native-like’ or ‘near native-like’ or a rigid sense of ‘fluent’, you should use the research term ultimate attainment to refer to the steady-state of overall mastery.
So why is comprehensible output the complement to input, and what does a multilingual perspective and fluency have to do with it?
Output and production are closely related. Production and generative use are aspects of output. Production can be used synonymously with output but also specifically refers to retrieving word forms, such as when we take recognition-oriented vocabulary cards in Anki (i.e. word→meaning) and flip them when they’re mature (as tagged by Morph Man or this add-on), doing them as production cards (meaning→word) thereafter. Generative use refers to using learned aspects of the language together in new ways.
These types of output-focused learning are more effective than simple receptive processes, i.e. producing a word or constructing a sentence is more effective than retrieving a word’s meaning or reading an example sentence (or reading 3 example sentences). The output process is operationally distinct from input, though language production and comprehension share resources and tendencies. Learning a vocabulary word by production, for example, makes you better at producing that word, and learning it by recognition makes you better at recognizing it. But production also enhances recognition so effectively that if you have to choose between one or the other, it’s better to choose production over recognition.
Remember that with learning kanji, kanji are the targets, so we retrieve those, not the keyword, which is meant to fall away. We don’t have to worry about this leading to poor recognition, because being able to produce them through mnemonics and spaced retrieval is so robust that instant recognition becomes second nature.
Input alone doesn’t lead to much productive ability, and it leaves many gaps in your learning.
Comprehensible output refers to pushing the learner to speak and/or write the language in a way just beyond (hence the +1) their knowledge. It needs to be meaningful and primarily contain language items you know well, and you scaffold your output with strategies/tools/resources to account for errors. This is another case where explicit learning comes in, used to bracket your meaningful practice.
Output practice promotes an independent kind of learning from input. It encourages a bottom-up, syntactic kind of processing. This rather than top-down and semantic input learning, where often getting the gist, glossing over errors, seems good enough, and frequently you don’t know what you don’t know. Output inclines you to notice the gaps in your knowledge, which you parlay into noticing how others accomplish the same task, thus improving intake/uptake.
Remember Seinfeld’s stand-up routine where he notes how when you’re moving, you start noticing boxes everywhere? It’s kind of like that. It encourages you to listen like a speaker and read like a writer. Output promotes proactive testing of your knowledge, where you’re putting it all together and bringing it to the fore, experimenting. It also encourages a kind of metalinguistic perspective, where you become more aware of how various language constructions are used to make meaning.
Sounds amazing, right? Just like fluency isn’t an ambiguous word referring to some late, perfected stage of learning, and instead refers to polishing your ability to use mature (mastered) pieces of the language, output just means taking those things and using them productively, at the beginning and onward. You can even combine them so your output practice, such as strategies described below, is under time pressure (via plugins), a central aspect of fluency development exercises.
How do we get started?
What I’m interested in are strategy types designed for the solo player, as the subtitle might have indicated.
Output in Anki
First off, we’re going to be using translations a lot, which if you’ve been sadly indoctrinated into certain antiquated models of language learning which ignore research, might disturb you.
As you should be using throughout your study until you become fairly highly proficient, we’re going to be using plenty of the L1. Actually, the L1 isn’t just fine as a beginner/intermediate learner, it’s equally effective at late stages. Translations in your best language are transparent windows onto the semantic units that new word forms refer to. A bilingual person isn’t two monolinguals in one head, the two languages interact automatically the instant you begin learning an additional language, extending a network that refers to the same semantic knowledge and relies on the same general language parsing abilities.
The languages thus interact, enhancing learning by allowing knowledge to be transferred, and necessarily creating a version of the target language that’s unique (e.g. rather than every difference from the native monolingual’s version of the target language being a mistake, deviating from native use, often it’s simply an artefact of multilingualism and bicultural identity). Translanguaging, using both languages, is a fundamental part of the multilingual’s identity and life, and being able to stick to one language as one desires comes from forming strong connections and building a repertoire, connecting new word forms to semantic knowledge, internalizing grammatical patterns, and using the L1 to progressively develop a strong L2 base through which using the L1 becomes merely an option, rather than a necessity.
So come out of the cocoon people online might have encouraged you to seal yourself into.
Type 1: o+1 Sentence Cards
You’ll recall that generative use refers to using familiar language items in novel combinations. In Anki, we’re going to use Morph Man* to find the sentences with 0 unmature words. We’re using i+0 sentence cards as o+1 production cards. How does this work, you wonder? I’ve talked about flipping cards at this all-mature stage, and using a different template (e.g. Production, or Comprehension) to study them. We’re not doing that here. These will be new sentences we’re using. Not too short or too long, and composed entirely of mature words**.
So here the +1 refers to the overall sentence structure. Ideally I would prioritize literally unfamiliar sentence patterns over familiar ones, but it doesn’t matter as long as the sentence is new.
As I’ve reminded you, I advocate doing vocabulary cards as recognition first, then switching to production when they’re mature. This offsets the difficulty of production cards. I’ve also noted we’re not doing that here, so that we can address novel sentences with mastered words. In part, the cohesive context of the sentences offsets the increase in difficulty that comes with sentence production cards. But we’re not stopping there.
We’re not going to produce these words based on a meaning (translation) cue, like a vocabulary production card with too many words. That’s not the goal of these o+1 cards. The goal is the structure, remember? So we’re going to use the sentence glossing plugin, and place not only the translation of the sentence, but the list of the sentence’s words and definitions on the front as well. I was so caught up in this unshuffling idea, I overlooked that the plugin lists the words in order, thus giving you the answer, so you might want to randomize the gloss field, as a certain programmer-god has pointed out. The glosses will be dictionary forms of primarily content words (nouns, verbs, etc.). Update: The Sentence Gloss Shuffle add-on now takes care of the randomization issue.
When grading, we therefore want to focus on inflections, particles, and word order, while of course considering the meaning that emerges from the Japanese construction. We should also be using sentences with audio or video (e.g. from something like Core 2000 or subs2srs), so we can work on prosody. For output prosody, we’re going to be trying to immediately speak the sentence aloud given the cues (rather than my past recommendation to subvocalize first for sentence input cards, flipping and only then listening to/repeating the target audio).
We’re working at a higher level of abstraction and sophistication here, with words we’ve already multimodally mapped to meaning (e.g. multisensory word form as cue, meaning as target, and vice versa, because learning words and their meanings with the senses combined increases retention of all elements); thus we’ll be seeing more overlap of patterns, and we can be flexible in our grading. If you miss just an inflection or particle, depending on sentence length that won’t be a big deal, so no need to fail (grade instead as ‘hard’, perhaps); for word order, that’s larger and more specific, so you should probably fail if you order a word wrong. For vocal prosody, there’s going to be even more overlap and generality here, especially as we learn to use our own voice, so no need to grade this, just be diligent and attentive as you practice.
(Not that you ever need to be too rigid with grading, at any level. This is the qualitative aspect of the semiautomatic Anki, your heterophenomenological life partner. Especially with items you’re familiar with, you will know how badly you failed and whether the corrective feedback during reconsolidation (modifying a memory after retrieval, when it’s plastic) was enough. Whether forgetting or misplacing that kanji component was a dealbreaker. If you’re doing a multisensory vocabulary card as production, especially if you learned it as recognition first, as I recommend, you only need to get the reading or writing correct. It’s natural to have one or the other unevenly mapped as you move on through Anki and outward into multidimensional usage in multiple contexts.)
So on the Front we have: Translation of the sentence; Glosses, and possibly a picture/written explanation of context; any notes, such as comments on style, register and politeness level (differentiating between PL2/PL3 is probably most important, e.g. だ vs. です, ある vs. あります; see Japanese the Manga Way), whether it’s a sentence more appropriate for written or spoken language, etc.
On the Back: Sentence, Reading/Audio/Video
Remember, the translation is just there to provide a sense of the meaning of the sentence. A transparent semantic guideline that gets subsumed by Japanese usage and feedback. Using the guideline to take the Japanese words and put the sentence together are the important things here, constructing a target Japanese utterance based on a communicative goal and a set of known words.
*Here’s a simpler explanation of Type 1, which doesn’t necessarily require Anki.
**If you’re feeling adventurous, feel free to use sentences w/ an unknown word. Just make sure to remove it from the glosses on the Front. Also, ‘not too short/long’ is obviously subjective. Possible sources could be the Core 2/6/10k decks, KO2k1, subs2srs, etc.
***Update: Years later, it occurs to me that if we have the shuffled glosses on the Front, there’s no real need to require the words to be known.
Type 2: Adjacency pairs and subs2srs
This is exciting. I’ve been obsessed with ‘imaginary conversations’ for years. Blame it on Landor. Anyway, I couldn’t figure out a solution to the inability to chain cards together into conversational batches to SRS.
Adjacency pairs and the context lines that subs2srs allows did the rest to aid constructing this method. An adjacency pair is a slice of conversational turn-taking, an utterance and a response. Speaker 1: How you livin’? Speaker 2: I am quite fine, thank you.
Take cards made with subs2srs. We will need at least and specifically 1 leading line, so ensure that’s enabled in the options. We are going to use, once again, sentence cards (preferably with video clips, for full multisensory integration, but audio is okay), with entirely mature words in the main expression. The main expression also needs to be preceded by a line with a different speaker.
The template will be designed so that we have the leading line expression (and if you want, the leading line’s meaning/media/gloss) on the Front. Our comprehension for this is not what we’re testing/grading on. That practice is just a nice bonus. Instead, given the leading line cue, we must come up with the appropriate response, which is the main expression of the card, placed on the Back. Therefore we also aren’t using glosses of the target expression on the Front, only the glosses of the cue. We don’t want hints as to the appropriate response. Thankfully with our turn-taking focus, we have relatively rich context to offset the increased difficulty of recalling multiple words. We can also focus on much shorter, common sentences if we like (you can control subs2srs card length).
Sometimes the target on one card will be a cue on the next card. This doesn’t matter, as we’re only interested in the turn-taking dynamic, and having the target as a cue on another card won’t tell us what the preceding line was. That particular pairing will only be on one card. Logically following from this is that we are not showing Trailing lines at all, and if we have more than one leading line, we are only showing Leading Line 1 (expression/meaning/etc.), on the Front.
You might also want to select cards more specifically for these adjacency pairs, that is, focus on sections of videos where there are interesting conversations, perhaps with particular speakers you want to emulate or styles you want to focus on, and limit the spans to those sections in subs2srs, only excising those.
Also note that many Japanese subtitles include speaker names in parentheses. This will cause Morph Man to think i+0 sentences are i+1, unless you’ve got the name in your known/mature database. You might want to select adjacency cards in the browser and in the expression field, find/replace （.*） with nothing, with regexp enabled.
Oh and don’t worry, since you’re not a primitive robot, you’re just internalizing possibilities for responding to dialogue cues in a variety of situations, based on what you want to say. Studying turn-taking this way lets you practice constructing meaningful sequences based on context and the perspective of a real or imagined speaker as part of a discourse.
Update: Another source of adjacency pairs can be found here. So you can simply reply to Speaker 1 as Speaker 2 for the ‘Pairs’ card type.
Type 3: Microblogging
Another source for adjacency pairs is Twitter. Twitter is more of a hybrid speaking/writing type place, with a sense of immediacy to interactions, and the medium inclines people to be brief both in utterance and in conversation length. Here are some tips for Twitter and Japanese.
There’s also simply using microblogs, with their brief inclinations, to reblog or write short comments. One thing that might work here is, take grammar patterns and/or vocabulary you’ve learned to maturity in Anki, and your task is to use a selection of this to compose novel, meaningful sentences to post on Twitter or tumblr. Something pertinent to your microblog persona. You can use various tools to check your output, such as JGlossator, or various corpora (Google, Anki decks, corpora mentioned on the tools/resources pages, etc.).
What might discourage you from this, and perhaps from output in general, is a misguided pressure to be like a monolingual native, to present yourself as having reached a certain level of competency or else you’re being fake and deceptive. The only thing you need to worry about is doing whatever you want and whatever it takes to get better at using the language, as yourself, an L2 user who has decided to become multilingual and thus every unique, incremental step on that path is an additive success. Only worry about your current knowledge in the sense that you want to continuously take that knowledge, from the onset, and work on making sure you can use it fluidly and productively.
Type 4: Grammar
You can also try clozing sentence structures that represent grammar points/patterns. Place the explanation on the front as a hint (assuming the explanation doesn’t explicitly give away the target structure, obviously). Again, do this only for unfamiliar i+0 sentences, as here the focus isn’t on vocabulary.
Miscellaneous
- In a sense, crossword puzzles are a kind of output practice. You can try using English definitions and/or Japanese sentences as clues.
- Back translating.
- For production cards, consider enabling the type-in-the-answer function, comparing with the target field. Use the honor system and don’t type it in until you have decided you know or don’t know the answer (e.g. the kanji in a word), because typing will bring up the candidates via the IME.
- Mad Libs. Use Morph Man to generate a list of words with their parts of speech. Find the words in the text using regexp, Replace with the parts of speech. So you fill in the blanks in the text based on the POS hints. You can study the text then try to fill in the missing words, arbitrarily fill in words of the same word class for humorous effect, ha, ha, or what I think would be best, find/replace only the known words of an unfamiliar text, and use the list of those known words to put them back where they belong based on hints and context. Compare with original unclozed text to make sure inflections are correct. Previous posts have described various ways of finding/replacing words in texts based on Morph Man lists.
- You might also try reassembling short text passages from unfamiliar texts consisting entirely or almost entirely of known items. Given the sentences, you must put them in order. You can use Morph Man on the texts, then a simple find/replace with regexp to break the sentences into lines, such as Find: ([！,？,。]) and Replace: \1\n – Then you can sort the sentences with your text editor or Text Mechanic. This will force you to be aware of text-level aspects of discourse, including cohesive ties. You might include 、 as a delimiter also. See also: 起承転結
- Other options include games like this free web-based Shiritori wordplay game.
- Don’t forget aizuchi (back-channeling), for discourse-level practice.
