Russian

Most language learners are familiar with Forvo, a site that allows users to download and contribute pronunciations for words and phrases. For my Russian studies, I make daily use of the site. In fact, to facilitate my Anki card-making workflow, I am a paid user of the Forvo API. But that’s where the trouble started.

When the Forvo API works, it works OK, often extremely slow. But lately, it has been down more than up. In an effort to patch my workflow and continue to download Russian word pronunciations, I wrote this little scraper. I’d prefer to use the API, but experience has shown now that the API is slow and unreliable. I’ll keep paying for the API access, because I support what the company does. And as often as not when a company offers a free service, it’s likely to be involved in surveillance capitalism. So I’d rather companies offer a reliable product at a reasonable price.

Welcome to Part III of a deep dive into my Anki language learning decks. In Part I I covered the principles that guide how I setup my decks and the overall deck structure. In the lengthy Part II I delved into my vocabulary deck. In this installment, Part III, we’ll cover my sentence decks.

Principles

First, sentences (and still larger units of language) should eventually take precedence in language study. What help is it to know the word for “tomato” in your L2, if you don’t know how to slice a tomato, how to eat a tomato, how to grow a tomato plant? Focus on larger units of language increases your success rate in integrating vocabulary into daily use.

In Part I of my series on my Anki language-learning setup, I described the philosophy that informs my Anki setup and touched on the deck overview. Now I’ll tackle the largest and most complex deck(s), my vocabulary decks.

First some FAQ’s about my vocabulary deck:

1. Do you organize it as L1 → L2 or as L2 → L1, or both? Actually, it’s both and more. Keep reading.
2. Do you have separate subdecks by language level, or source, or some other characteristic? No, it’s just a single deck. First, I’m perpetually confused by how subdecks work. I’d rather subdecks just act as organizational, not functional, tools. But other users don’t see it that way. That’s why I use tags rather than subdecks to organize content.¹
3. Do you use frequency lists? No, I extract words from content that I’m reading, that I encounter when listening to moviews or podcasts, or words that my tutor mentions in conversation. That’s what goes in Anki.

Since this is a big topic, I’m going to start with a quick overview of the fields in the main note type that populates my vocabulary deck and then go into each one in more detail and how they fit together in each of my many card types. At the very end of the post, I’ll talk about verb cards which are similar in most ways to the straight vocabulary card, but which account from the complexities of the Russian verbal system.²

Although I’ve been writing about Anki for years, it’s been in bits and pieces. Solving little problems. Creating efficiencies. But I realized that I’ve never taken a top-down approach to my Anki language learning system. So consider the post the launch of that overdue effort.

Caveats

A few caveats at the outset:

• I’m not a professional language tutor or pedagogue of any sort really. Much of what I’ve developed, I’ve done through trial-and-error, some intuition, and a some reading on relevant topics.
• People learn differently and have different goals. This series will be exclusively focused on language-learning. There are similarities between this type of learning and the memorization of bare facts. But there are important differences, too.
• As I get further and further into the details, more and more of what I discuss will be macOS specific. I’m not particularly opinionated about operating systems. And my preference has more to do with the accumulated weight of what I’m accustomed to and as a consequence, the potential pain of switching. In the sections that deal with macOS specific solutions, feel free to skip over that content or read it with a view toward thinking about parallel tools on whatever OS you are using.
• I use Anki almost exclusively for Russian language acquisition and practice. Of necessity, some particularities of the language are going to dictate the specific issues that you need to solve for. For example, if verbs of motion aren’t part of the grammar of your target language (TL) then rather than getting lost in those weeds, think about what unique counterparts your TL does have and how you might adopt the approaches I’m presenting.

We that out of the way, let’s dive in!

In my perpetual attempt to make my language learning process using Anki more efficient, I’ve written a tool to extract English-language definitions from Russian words from Wiktionary. I wrote about the idea previously in Scraping Russian word definitions from Wikitionary: utility for Anki but it relied on the WiktionaryParser module which is good but misses some important edge cases. So I rolled up my sleeves and crafted my own solution. As with WiktionaryParser the heavy-lifting is done by the Beautiful Soup parser. Much of the logic of this tool is around detecting the edge cases that I mentioned. For example, the underlying HTML format changes when we’re dealing with a word that has multiple etymologies versus those with a single etymology. Whenever you’re doing web scraping you have to account for those sorts of variations.

Dealing with PUNCT nodes in interlinear glossing.

Still thinking about interlinear glossing for my language learning project. The leizig.js library is great but my use case isn’t really what the author had in mind. I really just need to display a unit consisting of the word as it appears in the text, the lemma for that word form, and (possibly) the part of speech. For academic linguistics purposes, what I have in mind is completely non-standard.

The other issue with leizig.js for my use case is that I need to be able to respond to click events on individual words so that they can be tagged, defined or otherwise worked with. It’s straightforward how I could apply CSS id attributes to word-level elements to support that functionality.

leipzig.js is a library for applying interlinear gloss to texts for linguistic analysis. In this post, I experiment a little with this libary to evaluate whether it would work for a little project of mine.

Starting a new devlog about Hedghog, a new language learning app and some thoughts about the interlinear display of lemmas.

Splitting text into sentences is one of those tasks that looks simple but on closer inspection is more difficult than you think. A common approach is to use regular expressions to divide up the text on punction marks. But without adding layers of complexity, that method fails on some sentences. This is a method using spaCy.