Work predictably with large corpora
Speed depends on the size and structure of the corpus, the breadth of the request, and the amount of output produced. Exempla uses paginated tables, background work, cancellation, and result cleanup to keep large operations manageable.
Exempla has been tested with a corpus of about 10 million words
The reference test project contains 708 text files, 893 speakers, and 10,409,395 words when apostrophe forms are separated. On the development Mac used for the test, opening the project took about 3.1 seconds, Word List generation about 6.75 seconds, Corpus Statistics about 6.25 seconds, and a broad Contains search returning about 3.58 million results about 7.8 seconds.
Four factors matter most
More words, transcript lines, text files, and speakers require more reading, counting, and grouping.
Exact searches are usually narrower than Contains. A one-letter Contains query can return millions of occurrences.
The entire corpus normally requires more work than one text or speaker. Use an Active selection when the research question concerns a subset.
Generating, sorting, adding, or exporting millions of rows takes longer than showing one page of them.
Changing a view filter after generation can be quick, but it does not reduce the work already required to generate the underlying result set. Set the search mode, scope, apostrophe rule, case rule, sequence length, and minimum frequency before starting when those settings affect generation.
Old analytical results do not accumulate indefinitely
A new completed search replaces the previous Concordance result set. Clear removes the current result model.
Regenerating replaces the previous result for that workspace. Clear releases its generated table where a Clear command is available.
The Clipboard is intentionally different. It persists across searches and is saved with the project. Rows are stored compactly, with only the current page displayed.
After results are cleared, the operating system may keep reusable memory reserved for the application. The memory number shown by macOS or Windows therefore may not immediately return to its launch value, even though the old result tables have been released.
How each analytical workspace manages large output
Search work runs outside the main interface. Results are paginated at 1,000 rows. Broad result sets can still take time to generate or sort. See Large searches.
Generation counts the selected corpus scope. The Minimum control filters the generated list; changing the corpus or tokenization rule requires regeneration.
Generation counts consecutive word sequences for the selected length and scope. A large corpus can produce many unique sequences, so use a meaningful minimum frequency.
Generate or Refresh recalculates totals and complete text and speaker tables after relevant corpus changes.
Disk-backed storage and pagination keep the visible table compact. Very large collections can still increase project size and export time. See Large collections.
Progress means Exempla is still working
Long searches, generation jobs, large Clipboard additions, saves, and exports use progress or status feedback. Where a Cancel control is provided, use it once and allow the operation to stop cleanly. Do not repeatedly click the original command.
Review the result count and elapsed time shown in the page header, status message, or completion notice.
Cancelled search and generation work does not replace the last completed state with partial results. Large CSV exports do not leave a partial destination file.
Exports of 5,000 or more rows use a progress window and can be cancelled.
Use a small verification pass before a very large operation
Check the Active selection and the tokenization options that affect the result.
Use a narrower scope or a more precise query to confirm that the output has the intended meaning.
Generate, inspect the total, then export directly when you need the whole result set.
For hundreds of thousands or millions of Concordance results, direct CSV export is normally more efficient than placing every occurrence in the project Clipboard. Use the Clipboard when you need a reusable evidence collection inside the project.
If the application feels slow
The operation is active. Let it finish or use Cancel. Avoid starting the same task again.
Sorting a multi-million-row result by Speaker, Source, Token, or Line can require a complete reorder. Wait for the elapsed-time update before repeating the click.
Wait briefly, then use Cancel if available. If the interface does not recover, preserve the project if possible and restart Exempla.
Clear finished generated results and export then clear very large Clipboard collections. Closing unrelated applications may also free system memory.
Include measurable details in a performance report
Record the operating system, computer model, installed memory, Exempla version, project word/file/speaker counts, Active selection, exact query or generation settings, result count, elapsed time, and whether progress or cancellation remained responsive. Do not send confidential transcript content unless you have permission.
Was this page helpful?
Your response will help improve the guide.