Repository navigation
Wordflow 0.7.20
Wordflow 0.7.20 makes the tools simpler to start with and easier to read. Four tools gain an Advanced switch that keeps their less-used settings out of the way, any result or the Data Editor can be maximised to the whole window, and Trends can show rates per million words. Topic Modelling places more of your text in topics and no longer freezes on results with thousands of topics. Shared servers keep your AI providers and keys after a page refresh.
New in 0.7.20
Simpler tools, more room to read
- Basic and Advanced settings: Trends, Concordance, Annotation and Topic Modelling now show an Advanced switch at the top of their settings. With it off, you see only the settings a first run needs, and Wordflow uses the standard value for every hidden one, so a hidden setting never changes a result. Turning it on shows the rest; anything you set there comes back when you turn it on again. Wordflow remembers your choice for each tool on this computer. Frequency, Keyword Analysis and Quotation have no advanced settings, so they look as before.
- Where settings moved: in Concordance, Search mode, Ignore punctuation and the Combined view for two Data Blocks are advanced, and Use regular expression is now Regular expression. In Annotation, the example Data Block, the prompt and the inference and Run settings are advanced; the switch replaces the Advanced settings fold. In Topic Modelling, Passages, Max tokens, Seed, Topic sampling and Ungrouped passages are advanced. A tab whose last run used one of these opens with Advanced on.
- Maximise results: the new button on the bar between a tool's settings and its results gives the results the whole window. Maximise Data Editor, in the Data Editor's title row, does the same for the table. Restore layout or Esc brings everything back as it was, with your settings, selections and pane sizes unchanged.
Trends
- Rates per million words: Trends counts rows, which misleads for Concordance matches: more matches in a period may only mean more text was collected then. With Advanced on, Normalise against lets you choose a reference Data Block, by default the one your Data Block was made from, and Show as then offers Per million words and % of reference rows. A period with text but no matches shows as zero. If the reference lacks one of your Group By columns, a note under the chart says what the rates divide by.
- Show as replaces Normalise to 100%: the choices are Row count and Composition (% of period), plus the two rates when a reference is chosen. A tab saved with Normalise to 100% opens as Composition.
- Minimum group count explains itself: a ? beside the setting says what it does and opens its help.
Topic Modelling
- More of your text in topics: Topic Modelling often left more than half of a corpus Ungrouped. The new advanced setting Ungrouped passages lets left-out passages join the topic they are most similar to. Join a topic they fit admits a passage only if it is as similar to the topic as the topic's own passages are; in our tests it took one corpus from 58% Ungrouped to 18% with the same topics. Always join the nearest topic leaves nothing Ungrouped and is not recommended for most work. The topics, their words and the map are the same with or without joining, and joined passages are marked Joined in a topic's examples. The default, Stay Ungrouped, gives the same results as before. When 30% or more of your documents are Ungrouped, Wordflow now suggests this setting.
- Max tokens sets itself: Max tokens now starts on Auto. When you choose Run, Wordflow measures all the text and picks the passage size that keeps 9 in 10 paragraphs, lines or sentences whole, then shows it, for example "Auto (64)". On short texts this can give different topics from 0.7.19, which always started at 256. Choose a size yourself to keep it fixed.
- No freeze on results with thousands of topics: a low Min topic size on a large corpus can find thousands of topics, and opening such a result froze the app. Results now open with at most 100 topics, merged from the ones found; the Topics slider still reaches every topic.
Concordance and Quotation
- Concordance results are named after the search: a new Data Block or downloaded file is now called, for example, speeches_concordance_climate, so several searches of one Data Block can be told apart. You can still change the name.
- Clearer Concordance columns:
CONC_matched_textnow reads "Text found" andCONC_extractionreads "Match in context", each with a fuller explanation when you hover. The column names themselves are unchanged. - Quotation exports no longer repeat the document:
QUOTE_extractionwas a second copy of the whole document in a new Data Block or file, and it kept the long text in a download even when the text column was unticked. Add to Project and Download no longer offer it. - The Quotation table uses the full width: the quotation column no longer stops halfway across a wide results pane.
Frequency and Keyword Analysis
- Sorting by Overuse works: sorting the keyword table by Overuse listed keywords alphabetically. It now puts the keyword most overused in the Study corpus first; click again for the one most overused in the Reference corpus.
- The tokeniser's About link shows with two Data Blocks: the link beside Tokeniser model was hidden when two Data Blocks were selected.
- A corrected hint: the hint before the first run mentioned stop words, which only appear with the results. It now says so.
Data Loader and Data Editor
- Advice before uploading thousands of files: uploading more than 1,000 files is slow, because each is sent separately; about 26,000 small text files took more than an hour. Wordflow now says so first and advises compressing the folder into a ZIP, which uploads as one file and makes the same Data Block. You can still upload the files as they are.
- A bigger preview window: Maximise preview, beside the close button of the file preview and the Add windows, shows the table on the whole screen.
- Long values no longer take over the Data Editor: a column of long values such as web addresses is now limited in width, with Expand to see it in full, as long text columns already were.
Shared servers
- AI providers and keys survive a page refresh: on a shared Wordflow server, a provider added under Settings → Servers showed "Key saved" and was gone after a refresh, as was the LDaCA access token. They now stay in your browser, for your account only, until you delete them, and Settings says where they are kept. On a shared computer, delete your providers when you finish. The desktop app was not affected.
Good to know
- A Project saved with a Topic Modelling run that used the Multilingual model and Max tokens Auto does not open in versions 0.7.16 to 0.7.19. This matters only if you go back to an earlier version.
- Server operators: no settings change. The server image now uses polars-text 0.7.3.