cantodict-archive

Archive of the abandoned Cantonese Dictionary cantodict from https://www.cantonese.sheik.co.uk/dictionary/.

This is an attempt to save the cantodict dataset as the project seems abandoned. It is an incredible dictionary consisting of lots of works from many volunteers over years. Letting it bitrot would be a major waste.

Copyright and License

The code is licensed under CC4.0 attribution. The dictionary data is a scrape of public contributions where the most prolific author (bybell) that contributed over 53% of the entries in the dictionary has explicitly allowed the republication.

Furthermore, the active members of this project has forked a version into Cantoneseplus which intends to keep info public.

Given that Cantonese is a waning language and Cantodict provides a snapshot history of usage, it feels important to preserve it. That veteran members of the project believe there is not a problem with a scrape, I'm repulbishing the data here.

Installing the dictionary

Inside the output directory, there are 4 types of files:

• cantodict-* directories contain Kobo and Kindle friendly dictionary files with the cantodict data. Note the "yale" version only converted the initial romanization from jyutping to yale. Any romanization inside the defintiion is still jyutping as I couldn't find a fast way to separate those out to convert.
• *.json files include json dumps of the extracted content of the data directory.
• cantodict.sqlite is an sqlite database combining all the files. See TABLE_CREATE_SQL in json-to-sqlite.rb for a commented schema.
• cantodict.csv is a dump of cantodict.sqlite in csv format.

To install the dictionaries on Kobo or Kindle, you need to side-load them.

Dictionary variants.

I prefer yale romanization so I created a variant of the dictinoary where the primary romanization presetned is in yale.

The dictionary also has some set of incomplete items, and 205 items that were marked as "Vulgar" which some parents may not want to show to kids. Note...there are probably a number of vulgar terms that are NOT flagged, but this removes at least a large chunk of them.

If you're uncertain what variant to use cantodict-jyutping is probably the one you want, or cantodict-yale if you don't know any romanization system as Yale is more natually readable to English speakers.

Both the Kobo and Kindle versions of the dictionary should support looking up variants of characters/phrases includes simplified.

Installing dictinoary on a Kobo

For Kobo, either use the dictutil to copy the wanted variant of dicthtml-zh_HK.zip on to your device. Or to do it manually, copy it into the .kobo/dict directory on the device.

For Kobo, you can only install one version of this dictionary easily since the filename is based on locale. If you want to install two variants, you could fake it by renaming files to zh_TW, zh_CN, etc.

Installing dictinoary on a Kindle.

For a Kindle, it means copying the .mobi file for the variant of the dictionary you want into the Documents directory of your Kindle.

Running the scripts to regenerate the output data.

bundle config set --local path 'vendor'
bundle install
bundle exec ruby pages-to-json.rb  # Generates the json extraction.
bundle exec ruby json-to-sqlite.rb  # Puts it all in an sqlite DB
bundle exec ruby json-to-csv.rb  # Puts all json data into csv file. Has both jyutping and yale romanizations.
bundle exec ruby make-dict.rb  # Creates kobo file with all entries. Use -h to see entry filter options.
bundle exec ruby make-dict.rb  -j # Creates kindle file with all entries. Use -h to see entry filter options.

Tools created

1. pages-to-json.rb -- Takes downloaded webpages and outputs a json version of the entires.
2. json-to-sqlite.rb -- Takes the output json files for characters, compounds, and sentences and then loads them into an sqlite database.
3. make-dict.rb -- Takes the sqlite database and produces dictionary files of various formats. Initially .df files usable by dictutil to create Kobo dictionaries, and an opf directory usable by Kindle Previewer to create a .mobi file for use on kindles.

Data files

Data files are full screpes of entry detail pages and word list from the cantodict website (see Scrape Method).

Files are compressed using brotli. This yields significant savings over gzip so it was worth the more esoteric archival format.

Scrape Method

Cantodict looks to be a simple database frontend with REST style URL structre rendering characters, compounds, and sentences based on the primary key of the entry.

Entries are of the form http://www.cantonese.sheik.co.uk/dictionary/[type]/[id]/.

A character URL example: http://www.cantonese.sheik.co.uk/dictionary/characters/389/ for character number 389, which is 香.

A compound URL example is: http://www.cantonese.sheik.co.uk/dictionary/words/61652/ for compound number 61652 which is 鑒識.

Sentences URL example is: http://www.cantonese.sheik.co.uk/dictionary/examples/741/ for sentence number 741 which is 做咩你咁慢㗎?.

Getting the full list of is is a little harder. It is pretty much sequential (See Stats section below for totals of each) but there are holes. The site itself has some scripts that produce lists of all entries. Here are the relevant ones.

All Characters (there is no single query): http://www.cantonese.sheik.co.uk/scripts/masterlist.htm?level=1 http://www.cantonese.sheik.co.uk/scripts/masterlist.htm?level=2 http://www.cantonese.sheik.co.uk/scripts/masterlist.htm?level=3 http://www.cantonese.sheik.co.uk/scripts/masterlist.htm?level=4 http://www.cantonese.sheik.co.uk/scripts/masterlist.htm?action=unsure

All Compounds http://www.cantonese.sheik.co.uk/scripts/wordlist.htm?action=&wordtype=0&level=-1&page=0

All Sentences (notice level=-1 and wordtype=0). http://www.cantonese.sheik.co.uk/scripts/examplelist.htm?level=-1&wordtype=0&filter_by_audio=0&page=0

Scraping was done by walking through all of the lists to create initial data. After that, each individiual detail page is walked to fill in missing info. Then all every missing "id" from 1 to the last entry id for each type was queried and inspected for missing characters.

Cantodict Stats

Stats from when this was scraped

| Entry type | L1 | L2| L3 | L4 | L5 | L6 | Total | |Characters |71 |702 |1282 |2622 |698 |0 |5372| |Compound Words |642 |496 |3835 |44189 |10321 |1610 |60719| |Example Sentences |30 |223| |531| |722| |72| 0| 1548| |Examples with Audio |84| |184| |159| |6| 0| 433|

Num Editors: 81

Character pages.

The general structure is

//*[@id="container"]/div[1]/center/table[1]
<body>
  <div id="container">
    <table>...nav stuff..</table>
    <br>
    <div>
      <center>
        <center>..ad stuff.. </center>
        <small>..ad stuff.. </small>
        <table class="cardborder charactercard">
          <tr><td>
            <span class="cardjyutping">gau<span class="tone">2</span></span>
            <span class="cardpinyin">jiu<span class="tone">3</span></span>
          </td></tr>

          <tr><td class="wordmeaning">
            [1] nine [2] [Wu] unit for distance
              <!-- note this may have a flagicon for "Archaic Name Obsolete Place" -->
            <div class="charnotes">
              Hyperlinked blob of text explaining variants, etc. May link to other characters.
            </div>
            Default PoS: <img alt="noun">
            Additional PoS: <img alt="adjecetive">
            <div class="charstrokecount">Stroke Count: 2"</div>
            <div class="charlevel">Stroke Count: 1"</div>
            <div class="charradical"><span class="chinesemed"><a href="http://www.cantonese.sheik.co.uk/scripts/wordsearch.php?searchtype=8&radical_num=5">乙</a></span></div>
            <span class="typedesc">"This character is used in both Cantonese and Mandarin/Standard written Chinese."</span>
          </td></tr>

          <tr><td class="chinesebig">箇 / 个</td></tr> <!-- If simplified, it has a slash. 九 only has one codepoint. 后 / 后 has a repeat since it simplifies to itself. -->

          <span class="cd_similar">
             <a href="http://www.cantonese.sheik.co.uk/dictionary/characters/661/" class="linkchar">丸</a>
             <!-- repeated anchors with class=linkchar -->
          </span>

          <span class="credits">
          This word has been viewed 1 times since 30th Oct 2012, was added by sheik on 18th Mar 2007 21:22 and last
          <a href="http://www.cantonese.sheik.co.uk/scripts/editlog.htm?filter=九" title="See list of all edits for this entry">edited</a>
          on 10th Jul 2009 17:06
          </span>

          <tr><td>High res image + vocab list</td></tr>
          <tr><td>Sponsors</td></tr>
          <tr><td class="white cantodictbg1">
            <hr>
            <a href="href="http://www.cantonese.sheik.co.uk/dictionary/words/1400/"></a>
            <span>...bunches of spans for related compounds</spans>

            <br>
            <a href="href="http://www.cantonese.sheik.co.uk/dictionary/words/185/"></a>
            <span>...bunches of spans for related compounds</spans>
            
            [etc]
          </td></tr>

          <tr><td class="white cantodictexamplesblock">
            <div class="example_in_block">
              <a href="http://www.cantonese.sheik.co.uk/dictionary/examples/1046/">
                 ...
            </div>
            <div class="example_in_block">
              <a href="http://www.cantonese.sheik.co.uk/dictionary/examples/1252/">
                 ...
            </div>
          </td></tr>

        </table>
      </center>
    </div>
  </div>
</body>