Phrase Pump algorithm

I’ve been using Phrase Pump for about a month now. In general I find it very useful for listening practice, but I have some concerns about it.

It might be good for you to describe the algorithm used for selecting the words used to find the example sentences. (I’m a language tool software developer myself, so I like to understand the algorithms of the tools I use.) Your help pages are surprisingly sketchy.

My first observation is that although each time there are new words and phrases coming up, it seems it is staying on the same words for a long time. I haven’t measured it, but it seems a long time. Is there a way to manage what words are used? Is it pure spaced repetiion? In the settings, I see a “New” control, which defaults to 15 %. Is that for new sentences or new words from which the sentences are drawn?

The Phrase Pump front page shows I have 1630 words that I’ve marked to learn (I only recently started to accumulate them. Previously, before using Phrase Pump, I would accumulate them from 1 or more videos, color coded for multiple videos, download them for one video for study in my flash cards, and then delete them, so I can start fresh.) It also says I am learning 36 words (3 months), 2 words (week), 1 word (day). The “learning” seems to imply “in progress”. Isn’t that way too slow a rate?

For the listening practice, when I study a sentence, if I understand a sentence before exposing the written form and translation, it seems like there should be a “knew it” button that would tell the algorithm to bring it back less frequently. It seems it repeats sentences I understand too often.

I don’t understand how it picks a few sentences for filling blanks or translating mode at the beginning. I would prefer to have the option of all one mode at a time, or mixing them up. One bug is that if the autotranslation comes out as “-”, these sentences are not excluded.

That’s my feedback for now. But thanks for your efforts in developing such a great tool!

1 Like