2 ms·
Interesting point. As I noted elsewhere, I do agree that there is room for improvements in this space. I admit that my post is mostly fuelled by being annoyed w
by wodenokoto 11y ago
Interesting point. As I noted elsewhere, I do agree that there is room for improvements in this space. I admit that my post is mostly fuelled by being annoyed with their marketing copy.
I think you need to be more careful about what are useful words. I remember my Japanese text book introduced the word for alien fairly early, and I thought it was stupid. Once I came to Japan, I realised if I had learned that word, I could say more funny things in my very limited conversations.
When I was studying in China, we were introduced to the word for diarrhea and how to describe the texture of our stool within the first 2 months. Hardly top 1.000 words, but if you've ever lived in China, you realise it can come quite in handy!
Frequency lists are often heavily influenced by newspapers and that is probably not representative of the things you will need in daily life or conversations. I have wondered if subtitles are more representative of normal conversations, and there are some research into building corpus from that, but I'm not really sure how you test the utility of your corpus.
- syllogism 11y agoNewspapers are likely a bad choice, yes. I've thought about this a bit. Subtitles seem good, and there are corpora of casual speech. I plan to heavily bias towards verbs. Nouns are much more long-tailed, and their frequency is "bursty" --- if the noun is relevant to the situation, it'll be used a lot. So you can get them when you need them. Otherwise you can always use a pronoun, or be vaguer.