Anki & Language Tools

These are some notes on how I’m using Anki for language learning, and a few other language learning tools I’ve found. There are a million other posts about Anki online, and nothing here is particularly novel, but a friend was asking about some stuff so I thought I’d write up a few notes.

Anki

A lot of my learning centers around Anki, a spaced-repetition (SR) app. For background on SR and Anki see e.g. the Anki docs, gwern, refold. I like Anki because of how flexible, extensible, well-documented and cross-platform it is: it’s not language-learning specific and I’ve used it for many other things (exams, memorising phone numbers, memorising poems, memorising the NATO alphabet..).

Spaced repetition is a very efficient way of memorising stuff. It’s far from a complete language learning solution, but I think that, early on, brute-forcing lots of vocab is one of the best things to do.

It quickly gets you to a level where you can recognise topics and communicate simply about everyday situations, which is motivating. I think that core vocabulary matters a lot more than grammar for basic communication: you can convey intent well enough even if you get your conjugation or declensions wrong.

It also provides a necessary scaffold for more ‘passive’ & time-efficient learning methods such as listening to podcasts & watching videos. In Greek I found that the level of even language-learning-oriented podcasts was quite high, and I didn’t get much out of them without having a few thousand? ‘foothold’ words that allow you to roughly keep on topic and infer some of the rest. This idea is discussed in Refold.

Vocab Cards

Most of my Anki notes¹ are simple, single-word, foreign word + english definition pairs, with a card in each direction. For example:

η εθνικότητα ⬄ nationality
ανακρίνω ⬄ to interrogate, question
εύκαμπτος ⬄ flexible, pliable (capable of being flexed or bent without breaking)

These simple, ‘out-of-context word’ cards are my default for a new piece of vocab because they’re quick to create, particularly when using motî. ‘Out of context’ cards often seem to be frowned upon (as opposed to sentences / sentence fragments with some context), but for most words they get the job done and are worth it for how quick they are to pump out.

To seed my deck I started out by manually going through a few ‘most common words’ lists (example—there are lots of these, and the quality / methodologies can be dubious, but it’s just a starting point), and now I mostly add words on the fly as I encounter them (and decide that I actually want/need to memorise them!). Depending on the language, you might have more luck with pre-made vocab decks on AnkiWeb, e.g. this.

For many words, a single card is enough to quickly retain it. Others are harder: I’ll usually notice after a few days that I am repeatedly lapsing.

Sometimes, this is because I’ve ended up learning a few words at the same time that easily ‘conflict’ in my brain and I confuse them. These can be phonetically or visually similar (e.g. recently παραγωγή & προαγωγή), or related in meaning e.g. synonyms and antonyms. When I notice this happening, I try and avoid learning both at once by suspending one for a few weeks until I have learnt the other well.

Some words are just genuinely harder to retain: brains are weird! If I find myself struggling with a word, I try and do a few things:

Vocab Cards with Context

Learning vocabulary in context is a classic piece of language learning advice… and it works! I’ll create one or two extra cards that use the word in a short sentence or sentence fragment. I’ll also bring those cards to the front of the queue (on iOS: Browse ➢ Select Cards ➢ Actions ➢ Set Due Date ➢ 0 days).

For example, when learning the verb στήνω (to set up, to put up, to stand, to pitch), I also added a card to pitch a tent ↔ να στήσω μια σκηνή (see FrequencyMan for a way to see these cards in ‘dependency order’).

Here are some ideas for these cards that seem to help:

• mix up the tense, gender, sentence structure etc. This allows you to learn conjugation and grammar at the same time as learning vocab. It also makes my recognition of words more ‘robust’ (a bit like data augmentation).
• make them funny or absurd. This seems to make them more memorable.
• use the fact that your brain is good at ‘auto-completing’ and auditory trigger. Put a simpler word at the start of the phrase, that you can easily recall, and repeat the phrase out loud. Then, when trying to recall the harder word, remember the easier word and your brain will usually step through the sentence and recall the harder word. For example to learn the word ‘sleepy’: ‘Η Ντόρις ήταν πάντα νυσταγμένη’ (‘Doris was always sleepy’).
• make them about real people you know, or real situations. For example, maybe Doris is a real, sleepy, person in your life.

I’ve experimented with using LLMs to automate this stage, with a prompt a bit like:

Here are some words I am trying to learn in <language>.
I would like to create Anki cards that use these words in context, to make them more memorable.
I would like these cards to be short phrases or sentence fragments (one or two clauses), with varied structure, tenses, subjects etc.
Each card should use one or two of the words below, along with other words.
Keep the other words simple, i.e. make sure they are words that I probably know already so that I am only learning the new words.
Make the cards memorable and sometimes funny.
Try to create a densely-connected web of cards and vocabulary.
Keep the cards short—mostly up to 6 words (they need to be suitable for a flashcard).
Make sure the meanings are logical: don’t combine words artificially just for the sake of making a card.
Feel free to use words closely related to the below, e.g. past participles of verbs, or adjectives related to nouns, or adverbs from adjectives.
Use multiple cases: nominative, accusative, genitive.
Output the new cards in Anki CSV format with comma separators.

Then paste the list of words below the prompt. Anki has an automatic ‘leech’ tag for repeatedly failed cards, which is a useful filter. You can export note as plaintext (CSV) from the desktop Anki ‘Browse’ table (right click ➢ Notes ➢ Export). Then you can re-import from CSV.

The quality of these cards vary, and it is easy to over-do it and up with repetitive cards that waste time (which is why I haven’t automated it), but it can be a good starting point and you can manually pick and choose. You could fully automate this kind of thing with AnkiConnect.

Mnemonics

Each of my vocab cards has an extra ‘Notes’ field which is revealed when the card is flipped (in either direction). I use this field to store etymology, related words and other memory aids that can help me remember the card if I forget it.

Some examples:

• το κίνητρο: the motive, incentive. Notes: ‘the driver → αυτοκίνητο (the car)’ (these are etymologically related via κίνηση).
• χαραμίζω: to waste, to squander. Notes: ‘haram’ (the verb sounds like ‘haramizo’ and to waste is a ‘sin’)
• η απήχηση: the sensation, impression, the echo. Notes: ‘από + ήχος’ (breaking down the word: ‘from + sound’).
• το αφεντικό: the boss. Notes: ‘authentic’ (the Greek sounds like ‘afediko’. There is no logical connection here, but in my head I picture an ‘authentic’ gangster boss and it’s enough to trigger to association)
• διαπράττω: to commit (e.g. crimes). Notes: ‘prat’ (the Greek sounds like ‘diapratto’, connection is ‘someone who commits crimes is a prat’)
• η συντροφιά: the group, company (of people). Notes: ‘eating together’ (the συν- prefix means together, τροφή means food).

These notes don’t have to be etymologically sound, or logical. They just need to trigger a connection or mental picture. Often, the more absurd or tenuous they are, the better they seem to work.

Phrase Cards / Linking Words

Early on, I also added a lot of cards focused on ‘canned’ phrases, particular idiomatic ways of phrasing things, on definite/indefinite pronouns, and on ‘linking’ words—various prepositions and conjunctions.

These are very useful for communication and are I think are really valuable rote-learning as templates early on (even if you don’t understand the grammar e.g. because they use tenses you haven’t learnt). For example:

• ‘How would I say x?’
• ‘What does this word mean?’
• ‘I would like to go to x’
• ‘Despite X, I …’
• ‘Either X or Y’
• ‘Neither X nor Y’
• ‘Every so often’
• ‘When that happens, …’
• ‘When that happened, …’
• ‘Having done that, he…’
• ‘While I was doing that, …’
• ‘After that, …’
• ‘In the meantime, …’
• ‘Whose X are these?’
• ‘Who does this belong to?’
• ‘I do X while you do Y’
• ‘It doesn’t surprise me’
• ‘Some of the cats’
• ‘Those cats’
• ‘These cats’
• ‘Whose cats?’
• ‘The cat’s X’
• ‘The cats’ X’
• ‘Certain people’
• ‘Finally, we arrived’
• ‘I have no money at all’
• ‘It is under / on top of / next to / behind …’
• ‘At 9 in the morning’
• ‘As soon as …’
• ‘From now on …’
• ‘For the first time …’
• ‘Just as I started, …’
• ‘By the time X happens, …’
• ‘Not yet’
• ‘Already’
• ‘Since then, …’
• ‘At the same time …’
• ‘The day after tomorrow / day before yesterday’
• ‘On the other hand …’
• ‘In order to [verb]…’
• ‘As long as …’
• ‘Even if …’
• ‘Instead of X’
• ‘Regarding X’
• ‘Except for …’
• ‘Because of X …’
• ‘That’s why …’
• ‘According to X, …’
• ‘Most of the X…’
• ‘The majority of the X’
• ‘Not enough X’
• ‘Too much X’
• ‘Too many X’
• ‘A few of…’
• ‘More than / less than’
• ‘As [adjective] as possible’
• ‘The same as’
• ‘Different from X’
• ‘Across from …’
• ‘In the middle of …’
• ‘On the corner of …’
• ‘All the way to …’
• ‘Close to …’
• ‘Far from …’
• ‘Somewhere else …’
• ‘Everywhere / nowhere / anywhere’
• ‘Everyone / no-one / someone / anyone’
• ‘I mean X’
• ‘Actually…’
• ‘Essentially, it’s …’
• ‘By the way …’
• ‘Speaking of X, …’
• ‘As I was saying, …’
• ‘In my opinion, …’
• ‘I don’t know, but …’
• ‘What do you mean by X?’
• ‘Can you repeat that (more slowly)?’
• ‘I used to [verb]’
• ‘I have always [verb]’
• ‘It’s possible that …’
• ‘It’s probable that …’
• ‘It’s unlikely that …’
• ‘I’m supposed to …’
• ‘I should have [verb]ed’
• ‘I wish I could’
• ‘I suppose I could’
• ‘It depends on …’
• ‘It wonder if …’

These tend to show up organically when trying to communicate (and a good way in general to practice, and to find blind-spots, is to role-play situations / dialogues in your head), but LLMs are also quite good at coming up with them in bulk.

Conjugation

Anki is very good for learning conjugation.

Conjugation is often presented in terms of verb groups. I think that’s useful as a reference, but for speech you need to know things instinctively. For that, I’ve preferred to just take a shotgun approach and brute-force learn the conjugations of many verbs. Your brain will automatically compress the patterns and groupings!

There might be some pre-made decks in your language. In Greek, I worked through some of this deck, though it ended up quite tedious because of how formulaic conjugation is (having a card per verb+tense+person means most card are trivial, but they still slow down reviews). I suspended most cards and only left one card per verb+tense+person except for irregular ones.

I also have many cards that are simple subject + object + verb sentences from native → target language, with a range of tenses and persons. I prefer these to the conjugation cards from the pre-made deck, because you spend less time mentally parsing the instructions (‘to hit: second person singular future simple’) and just directly translate from the auto-brain-painted native language sentence (‘you will hit’). LLMs are good at generating piles of these sentences.

Grammar

I’ve got at least one card for every core grammatical setup (basic sentences with every combination of case, plurality, gender etc). See How To Learn German Grammar With Anki for some ideas.

Failure Modes

Anki definitely requires some discipline and self-honesty to be useful.

A trap I sometimes fall into is mentally short-cutting a card, because I rely on the visual shape of a word or card, or the presence of a particular secondary word, instead of actually learning the word. I end up mastering the card, but don’t recognise the word in the wild when the context is altered or e.g. the declension is different.

This also happens when learning several new cards at the same time: instead of actually learning the cards, sometimes I only learn to distinguish them relative to each other within the context of that Anki session / few days. Your brain is very good at being lazy and overfitting!

The ‘data augmentation’ idea from above helps in these cases, especially if those cards get spread over time. I also think that, generally, trying to be as attentive and conscious as possible when reviewing helps (rather than just letting the cards glaze over you). I find this much easier in the morning. Speaking cards out loud also helps with engagement, and with pronunciation.

Text-to-Speech

Having foreign words / sentences automatically read out when I see them is helpful for pronunciation and engagement.

Anki has an inbuilt TTS facility in its template language, that uses the OS’s TTS service. It’s served my needs well—I mainly review on my phone, and modern iOS TTS is very good.

I’ve also seen people talk about the HyperTTS add-on. When ran, it uses various external services to generate sound, and embeds them in the notes. It looks very customisable, but I haven’t felt the need yet to add an extra step.

Anki Settings

I recommend removing the maximum reviews/day limit (i.e. setting it to 9999), so as to not artificially delay cards past their scheduled sweet-spot.

I use FSRS, which is a newer, smarter space repetition algorithm. It is worth reading about, and in particular understanding the role of the tuneable ‘Desired retention’. Making this number lower than the default of 0.9 will probably lead to a more time-efficient steady-state (in terms of time spent per retained card), at the expense of retaining fewer cards. For language learning, I think efficiency matters more than perfect retention. You can see some stats by clicking on the FSRS ‘Help Me Decide’ button in the deck settings:

Anki Add-Ons

The desktop Anki client supports Add-Ons. I mostly use Anki on my phone, but I sync between the phone client and the desktop client using AnkiWeb so that I can run these add-ons:

• FSRS Helper. This provides a few extensions to the core FSRS algorithm, including ‘balancing’ out cards (jittering them back and forward by a few days) in order to make your workload more uniform. If you’re reviewing on your phone, in theory you should sync and run this daily for it to work optimally.
• FrequencyMan. This automatically orders your new cards by word frequency and familiarity, so that you learn more common words first and that your cards build on one another. It has a lot of neat settings but I am using the defaults. I don’t think the word frequency list in Greek is very comprehensive, so I might poke around a bit with this.²

Dictionaries

motî / Wiktionary

motî is a fully-offline, multi-lingual dictionary app, with data from English Wiktionary.³⁴ It’s really fast and easy to use. The developer, Jan, is responsive and has fixed a few bugs that I’ve reported. It’s also got an exemplary privacy policy. Unfortunately, it’s iOS only.

One of its best features is the ‘Add to Anki’ button, documented here. It allows me to look up and add words to Anki in seconds, as they come up in conversation & class. I’ve got the following card templates set up that add an article to nouns:

{% if entry.partOfSpeech|lowercase == "noun" %}
  {% if entry.genders|join:"" == "m" -%}
    ο
  {%- elif entry.genders|join:"" == "f" -%}
    η
  {%- elif entry.genders|join:"" == "n" -%}
    το
  {%- elif entry.genders|join:"" == "mf" -%}
    ο/η
  {%- else -%}
    {{ entry.genders }}
  {%- endif %}
{%- endif %}
{{ entry.headword|to_html }}

{% if entry.senses.count == 1 %}
  {{ entry.senses|first|to_html }}
{% else %}
<ol>
  {% for sense in entry.senses %}
  <li>{{ sense|to_html }}</li>
  {% endfor %}
</ol>
{% endif %}

Another neat feature of motî is that it also searches the Wiktionary ‘pronunciation’ renderings of words, which helps if your spelling is bad or if you aren’t as fast with a foreign alphabet keyboard.

WordReference

WordReference has excellent online bilingual dictionaries, generally with more example sentences and subtle meanings than Wiktionary. Unfortunately, it is proprietary and can’t be consumed offline (and a lot of my language practice happens on the tube…). As a result, I don’t have such a slick flow for entering looked-up words into Anki beyond copy-pasting. There is a userscript that works on desktop—it would be nice to make it play well with AnkiMobile.

Podcast / Video Transcription Apps

Consuming (hopefully already interesting) video and audio content in your target language is an engaging, semi-passive way to learn, and see the language in context. For Greek see podcasts & video.

Now that speech-to-text has got pretty good, there are lots of apps popping up that will show a transcription overlaid on a podcast or video, allow you to tap to look up words, and replay tricky segments.

For podcasts, I have tried:

• LangTurbo
• Aisten.
• LingQ

They’re all pretty good, with slightly different user interfaces and different ways they try and lock you in. They all have free trials. I’ve stuck with LangTurbo.

For the same idea applied to video, I’ve recently started with Lingopie. It seems to have a reasonable amount of Greek content. I think there are plenty of similar apps out there. It’s a very effective way of learning a language.

AI Language Tutors

There is also a flurry of AI language tutors being built. I tried Langua for a bit, and was seriously impressed by the quality. I didn’t stick with it though: I just found the experience a bit odd. Making small talk with an AI feels awkward and meaningless: I don’t really want to tell it about my life. Perhaps if I didn’t have native speakers in my life that I could talk to, I’d feel differently. Will revisit.

Translation Tools

For a classic ‘type in a box and see a translation tool’, I mostly use DeepL nowadays. The translations are good, but the feature I really like is the ‘Alternatives’ button that shows other sentences or words you could use.

Clozemaster

I was paying for Clozemaster for a while. It’s not bad, but the sentences definitely have a ‘scraped data’ feel and are a bit repetitive and un-engaging. It felt like a less good use of time than e.g. doing Anki with more carefully curated sentence.

Other Resources

I like much of the approach presented by refold.la. I haven’t tried any of their language-specific courses, but the general guidelines laid out here contain lots of good ideas, and are refreshingly honest (it’s going to take a long time, mainly do Anki and consume content in the language).