When the Inductive Method Fails: Learning Very Complex Grammar with Anki
In my last post on language learning with Anki, I argued that most pieces of grammar are easy to pick up and solidify via inductive flash cards and spaced repetition.
But in this post, I want to talk about how I’ve seen this inductive method fail, and where its limitations are. Specifically, Anki puts us at risk of slipping into rote memorization in cases where we really need to shore up our knowledge with some deeper understanding.
To be clear, I’ve rarely had this problem. In most cases, an inductive, sentence-based approach to language study has been more than sufficient for me to master grammar in Spanish, Korean, and Ancient Greek. I just read the occasional textbook explanation, make a huge number of cards, and the grammatical tables and rules largely take care of themselves. I love this, because some grammar rules are obnoxiously complex when you write them out, and if I can avoid memorizing them directly and move them straight into “muscle memory” instead, it’s a win!
In practice, this means that my eyes kind of glaze over while reading grammar discussions or tables in textbooks. “Yeah, yeah, interesting, but I don’t know how to learn this. Show me some examples! I can Ankify those!”
But once in a while I hit an impasse with this strategy.
When Flash Cards Fail
Old English, of all languages, is what stymied me. How hard can it be?! It’s English!
The basic problem is that flash cards work best if I already understand the grammar well enough to know why (or most of why) I got a card wrong. And with Old English, I found that I often failed to understand why I was making mistakes.
Here’s how we want Anki to work:
- Anki: How do you say “[verb] a dark site” in Old English?
- Me: Uhh… feal…. fealwa stow?
- Anki: Psych! It’s fealwe stōwe
- Me: oh, of course, because stōw has got that long vowel and is a feminine noun, so it takes an -e in the accusative case; and fealu is one of those rare u-stem adjectives, and since adjectives agree with nouns in number and gender and case, we also add a -e onto the end to make it singular feminine accusative, but that combines with the special u-stem to create a “we” sound instead, fealu + -e = fealwe.
- Anki (10 minutes later): So how do you say “[verb] a dark site”
- Me (confidently): fealwe stōwe!
- Anki: cha-ching!
Old English is an easy language in many ways if you already know English. Conceptually it’s nouns and adjectives are not that different from declension in other languages like Latin or Greek. But I find them harder to learn than Latin or Greek, because OE declension hinges entirely on teeny-tiny vowel differences at the end of the word.
In practice, here’s how my daily conversation with Anki actually goes:
- Anki: How do you say “[verb] a dark site” in Old English?
- Me: Uhh… feal…. fealwa stow?
- Anki: Psych! It’s fealwe stōwe
- Me: oh. okay.
- Anki (10 minutes later): So how do you say “[verb] a dark site”
- Me: Um. Right. Um. fealwe stōw?
- Anki: no, I told you, it’s fealwe stōwe! Are you even paying attention?!
The basic problem is that flash cards — as their critics have tried to remind us time and time again — encourage me to rely on memorization even in cases when I shouldn’t.
Much as we like to make fun of the school of ye olde pen-and-paper translation, the truth is that people who study textbook exercises by hand instead of spaced repetition don’t have this problem. If you are stuck on a reading or grammar exercise in a book, the natural response is to look up the grammar that you are missing — scrutinize the tables, read the rules — until you can give the right answer and understand why you gave it.
The pen-and-paper crew doesn’t throw up their hands and say “whelp, I failed to answer within 15 seconds, guess I’ll just look at the answer and hope for better luck next time.”
Why Old English is Hard
If you don’t care about Old English, you might want to just skip down to my solution below. tl;dr: Old English declension is complex and confusing in a way that, say, Greek declensions are not.
Something about that “-we” ending on fealwe, and -u and -e endings more generally, is just complex enough that inductive sentence mining and harvesting examples from my textbooks — while just occasionally glancing at a grammar table — wasn’t quite enough to make me clear on what we are actually doing here.
This caught me off guard, because other declensions and semi-regular vowel changes in the past haven’t bothered me. Greek’s -ην ending for feminine accusative is easy to remember, for instance, and Spanish’s frequent o →ue sound change (poder → puedo), was easy to pick up just from seeing lots of examples — I’ve never had to agonize over that rule.
Part of the problem in the fealwe stōwe example is that to really benefit from Anki practice here, I need to understand several things, not just one or two:
- stōw is a typical feminine noun (known as the ō-stem declension)
- These ō-stem nouns take a -e suffix in the accusative
- fealu is a strong adjective here (as opposed to a weak adjective), and thus should also be declined as a feminine accusative in this instance
- The feminine accusative ending for a strong adjective is also -e (it’s similar to the ō-stem nouns in this regard)
- The -u in fealu is not a suffix — so we’re not talking about the neuter plural -u here or other uses of -u — but rather is part of the stem: it’s one of the rare class of u-stem adjectives
- So it keeps the -u, which combines with -e to become -we
This complexity by itself is arguably manageable. Greek poses similar challenges, after all.
I think what really makes Old English tricky for inductive learning is the redundancy in those little vowel endings (like -e, -u, and -a). Each of these vowels does a lot of work. They are reused for many different purposes: the same ending appears on several different cases and genders, differs in usage on nouns vs. adjectives, etc.; so they mean completely different things on different words. As a result, this pattern felt largely random to me for several months, and I had a really hard time mastering noun declensions via example sentences alone.
The -u ending in particular became my nemesis as an Old English beginner. Trying to predict when it would and would not appear in one of my example sentences was maddening.
- Hēo ne is stuntu (“she is not stupid”): here “-u” indicates a singular feminine nominative noun
- ongēan ðā scipu (“toward the ships”): here it’s a neuter plural accusative.
- fealu land (“a dark land”): as we’ve seen, fealu just happens to be a rare adjective whose stem ends in -u, in this case a singular neuter (not plural!)
- fealu stōw (“a dark site”): oh, but it also appears in the feminine singular, remember?
- gram bearn (“an angry child”): sometimes there’s no vowel on the words at all
- gramu bearn (“angry children”): other times we get a “-u” on the plural adjective, but nothing happens to the plural noun
The upshot is that all these various uses of simple suffixes encourage me to fall back on rote memorization, and I struggle to see the higher-level pattern that governs declension in OE.
Learning Grammar Explicitly
All my problems went away when I made a simple commitment, albeit one I have avoided for years in my language cards:
- When I have problems, I will study Old English grammar with Anki the same way I would any other scientific subject.
Now, we have to be careful here. I’ve made mistakes before when trying to do grammar in Anki. Here’s what doesn’t work:
- Making cards asking “what is the feminine genitive ending for ō-stem nouns?” Ew. I hated doing these cards.
- Making cards asking you to “rapidly chant” a declension drill (like students in a stereotypical Latin classroom). I appreciate this device in lectures, but it’s awkward in Anki, and has caused some leech cards for me.
- Making cards asking you to “write out” the whole declension table. This is a big, unwieldy exercise, and won’t schedule well with spaced repetition.
- Making cards that ask “why does such-and-such a word have such-and-such a form in the following sentence?”
These failures are exactly what drove me to the inductive method. But they are also not how I would approach “any other scientific subject.”
To study mathematics, chemistry, or algorithms, a good Anki approach is to search for big picture landmarks to orient myself, and then “hang” carefully chosen and motivated details off of those landmarks.
Here’s what that ended up looking like for Old English. In this case, Wikipedia ended up being a better source of landmarks than my textbooks, but sometimes it’s the other way around.
Big-Picture Landmarks for Nouns
Big-Picture Landmarks for Adjectives
Detail Cards
I’m not going to memorize the all details of all these tables just yet. Some parts I already find easy (like the -um ending for plural dative, which is unique and easy to recognize), and some parts I haven’t started learning inductively yet.
All I’m going to do right now is zoom in on the suffixes that I know have been confusing me: that pesky -u!
These cards don’t cover every possible use of -u in Old English (there’s still those rare u-stem words that have -u baked into the stem, for instance), but it’s been enough to clear up my confusion and let me get back to inductive language learning.
The fact that my detail cards refer to the landmark concepts (ō-stem and a-stem and strong adjective declensions) serves a double benefit: it gives me a way to target my prompts for detailed questions, but it also exercises the landmark concept in a different way. In general, when a landmark appears in many different cards from different angles, it becomes more familiar — easier to remember and reason about.
Conclusion
Overall, explicit grammar cards are rarely needed. Inductive practice with sentences will get you a long way. But a few carefully crafted grammar-overview cards can do wonders to unclog especially complex topics.
In this case, I hit a complexity barrier with Old English — there were just too many little bits of grammar I had to keep straight to answer simple cards. Putting some effort into grammar study reduced the complexity back to a manageable (but non-zero) level, so I can start making solid progress again by harvesting example sentences.
Ankifying the grammar knowledge is what worked for me, because I’m already accustomed to Ankifying scientific material. I considered other options, though:
- I could discipline myself to do Anki reviews in front of my textbook, and stop to read if I get confused.
- I could start adding some written translation exercises to my study habits, keeping a textbook handy for questions.
- I could just read a whole lot about grammar in one batch, and hope it sticks well enough to ease my Anki problems.
In my case, using Anki to solve my Anki problem felt easiest and like the most bang for my buck — but you might try one of these other strategies if you find yourself slipping into rote memorization!
