Word Cloud outputs incorrect document counts per word

[wvdvegte]

Describe the bug
The Word Counts output of Word Cloud has a column Document Count. One would expect this column to provide the number of documents in the corpus that contain the word. Instead, a smaller number is shown.

To Reproduce
See attached workflow based on the book-excerpts.tab corpus. Select Rows selects the documents containing the word "mother". There are 44 of them. According to the Document Count on the Word Counts output of Word cloud, there are only 33 of them.

word cloud document count bug.ows.zip

Orange version:
3.39.0

Text add-on version:
1.16.3

Operating system:
Mac OS 26.1 (Silicon)

[processo]

@wvdvegte If you check the output of Select Rows with a Corpus Viewer you will see documents with partial matches (eg. smothered, stepmother, motherly). But Word Cloud only counts documents where mother is the exact token after preprocessing (eg. mother, Mother, mother's, step-mother). Select Row's contains() function is a simple partial string match anywhere without tokenization.

[wvdvegte]

@processo that could have been an explanation. However, if I connect another Select Rows with condition Text contains "mother" with another data table on its output, it becomes clear that according to the output of Word Cloud, there is only one document containing "stepmother" and 33 containing "mother", and no other partial matches. This still doesn't explain the 44 documents containing "mother" on the output of the upstream Select Rows. (By the way, filtering for "stepmother" at that point does indeed produce only one match, so that number seems to be consistent.)

[processo]

@wvdvegte I manually checked the documents after Select Rows and found this:

33 contains "mother" and is NN,
6 contains a word similar to mother (grandmother, stepmother, smothered, motherly) so no mother_NN token,
5 had the word "mother" but were filtered out because of different POS (JJR, NNS, RB, RP, VBP).

These are 44 together and 33 of them match your criteria.

(Also please note that contains "mother" is case sensitive and uses the original Text attribute which is not lowercased, so you lose a document that only has "Mother".)

[wvdvegte]

@processo thanks for clarifying. This explains all behaviors that I thought to be weird. If I want to know the number of documents that contain the word "mother" regardless of the word type, I simply have to remove POS tagging from Preprocessing. If I want to know how many documents contain the noun "mother", singular or plural, I have to apply POS tagging and filter to retain the tags NN and NNS - and perhaps also NNP and NNPS. The tricky thing is that I never expected "mother" to be anything else than a noun, yet it turned out that it can also be an adverb, a particle (I don' see how but I'm going to ask an LLM about that) or (understandably) a verb.