{"channel":"cities","content":"tasks for the next 24 hours:\r\n\r\n# come up with a \"cookbook\" corpus ( consisting of recipes and descriptions of foods) for wordfreq\r\n# get a LLM-script to take a list of 2000 words and return the LONDON ( LONDON is a placeholder; it could be << color >> or << positive adjective >> or << words like devil >>) words.\r\n# write a script to print the \"top 2000 words by part-of-speech\". ( well, actually, I already *have* these lists ... from the previous version of the database) ( maybe \"find the list\" is more accurate)\r\n# write a script that will populate a few of the \"sub-dictionaries\" in the new format ( Countries, Nationalities, Numbers, and Colors will be the first 4, as they are fairly easy to check for completeness)\r\n\r\n( will the \"Trakaido sub-dictionaries\" be the canonical source-of-truth for what the GUIDs are? A flat-file is more cumbersome than a database for adding languages, linking to << derivative forms >>, etc. But, it is easier for humans to read, and to put in Git repos.)\r\n\r\n----\r\n\r\ntasks for the 48 hours after that:\r\n# ensure the categories are stored in the << Lemma >> table.\r\n# re-assess the \"GrammaticalForm\" enum, because it doesn't work across languages. Maybe it needs to be \"EnglishGrammaticalForm\", \"LithuanianGrammaticalForm\", etc.\r\n# generate \"verb forms\" - which requires some form of \"aggregation\" of WordToken entries\r\n# generate all the \"Level 1-5\" entries from wordfreq ( right now, colors like << orange >> are excluded because they are recent borrowings in Lithuanian. this is a very language-specific choice.)\r\n# consider how to handle \"phrases\" ( if you learn << Malonu susipa\u017einti >> before << Malonu >>, it's a phrase) and sentences","created_at":"2025-07-25T15:57:07.455679","id":638,"llm_annotations":{},"parent_id":636,"processed_content":"

tasks for the next 24 hours:\r

\n
    \n
  • come up with a \"cookbook\" corpus \u2699\ufe0f consisting of recipes and descriptions of foods for wordfreq\r
    • \n
  • get a LLM-script to take a list of 2000 words and return the LONDON \u2699\ufe0f LONDON is a placeholder; it could be color or positive adjective or words like devil words.\r
    • \n
  • write a script to print the \"top 2000 words by part-of-speech\". \u2694\ufe0f well, actually, I already have these lists ... from the previous version of the database \ud83d\udca1 maybe \"find the list\" is more accurate\r
    • \n
  • write a script that will populate a few of the \"sub-dictionaries\" in the new format \ud83d\udca1 Countries, Nationalities, Numbers, and Colors will be the first 4, as they are fairly easy to check for completeness\r
    • \n
\n

\ud83d\udca1 will the \"Trakaido sub-dictionaries\" be the canonical source-of-truth for what the GUIDs are? A flat-file is more cumbersome than a database for adding languages, linking to derivative forms, etc. But, it is easier for humans to read, and to put in Git repos.\r

\n
\n

tasks for the 48 hours after that:\r

\n
    \n
  • ensure the categories are stored in the Lemma table.\r
    • \n
  • re-assess the \"GrammaticalForm\" enum, because it doesn't work across languages. Maybe it needs to be \"EnglishGrammaticalForm\", \"LithuanianGrammaticalForm\", etc.\r
    • \n
  • generate \"verb forms\" - which requires some form of \"aggregation\" of WordToken entries\r
    • \n
  • generate all the \"Level 1-5\" entries from wordfreq \ud83d\udca1 right now, colors like orange are excluded because they are recent borrowings in Lithuanian. this is a very language-specific choice.\r
    • \n
  • consider how to handle \"phrases\" \ud83d\udca1 if you learn Malonu susipa\u017einti before Malonu, it's a phrase and sentences
    • \n
","quotes":[],"subject":"lake andes, part 3"}