The Dictionary: Get Technical Terms and Proper Names Right
An AI-supported transcription is a predictive process in which individual words can be spelled incorrectly. That makes a simple and time-efficient review of the transcript all the more important, which is why we provide the editor inside the program. Transcription errors can also be reduced in a targeted way, especially with the dictionary function. In specialist conversations and interviews in particular, technical terms can come up that the AI has rarely heard before and therefore finds hard to recognize. Giving the AI such words in a dictionary can support recognition considerably. This article presents the setting of this function and also its limits.
Create a dictionary
Under Settings > Dictionary you can manage your dictionary. Enter the technical terms that the AI should recognize. This includes rarely used or specific terms (for example a place name such as Heilstadt) and abbreviations (for example StPO, which the AI might otherwise hear as “Ess Tee Perso”).
The dictionary has two types of entries: a single technical term, which the AI receives as a hint, and a correction pair. A correction pair consists of the incorrect spelling and the correct spelling. After the transcription, memozero replaces only the incorrect spelling with the correct one.
The AI then takes all terms that you have entered in your dictionary into account during transcription. If new word errors or mix-ups appear during transcription that you want to take into account in the future, you can add them to the dictionary directly in the editor. The next section explains this.
Add words from the editor
If you notice incorrectly written terms in the transcript, you can take them into the dictionary directly. There are two ways to do this.
If you want to remember a single word as a technical term, proceed as follows:
- In the editor, change a word that is spelled incorrectly.
- Right-click the new word.
- Click Add to dictionary.
This path does not remember the incorrect transcription. The AI receives only the new word as a hint. That can still be useful if you want to prevent a specific error (for example “Krypto” instead of “Kripo”). Recognition sometimes softens crude expressions or replaces them with asterisks.
If you want to remember a correction pair, proceed as follows:
- Change one or more words in the editor.
- In the toolbar, click Remember changes. You find the button between the audio track and the text as soon as you have changed something.
- Select the change pairs that you want to take over.
- Confirm with Add.
In the dialog, nothing is preselected. You check yourself what memozero must remember. Only common errors belong in the dictionary, not one-off errors. Every word or word pair in the dictionary applies to all further transcripts.
memozero decides the type of entry on its own. If the two versions differ by more than capitalization, a correction pair comes up. If not, a technical term comes up.
Maintain the dictionary
Open Settings > Dictionary. The page has the heading “Terminology management and custom vocabulary” and the subline “Define technical terms and proper names to increase transcription accuracy”.
Caution: The trash can deletes the entry immediately. memozero does not ask you first.
| Element | Effect |
|---|---|
| Search field | filters by word and correction |
| Sort button | sorts A to Z and back |
| Add word | opens the dialog for a new entry |
| Pencil per row | changes the entry |
| Trash can per row | deletes the entry without a confirmation |
| Pagination bar | appears at 25 entries and more |
The dialog has a switch Correct a misspelling. It is off by default.
- Off: You fill in one field. A technical term comes up.
- On: You fill in two fields, Misspelling and Correct spelling. A correction pair comes up.
Enter confirms the dialog. The button stays gray while a required field is empty.
Limits of the dictionary function
The dictionary holds 500 words at most. The Remember changes dialog takes over 50 entries at most at one time. The keyword list that memozero sends to the AI is limited to about 900 characters; if not everything fits, memozero takes the most frequently used words first.
Transcripts that you have already created do not change when you add to the dictionary later. Changes in the dictionary can only be taken into account for new transcriptions.
When you add words, please also keep in mind that memozero can replace whole words only, never parts of words.
There is one dictionary for all transcripts; separate dictionaries per project or per client are not available. There is no import or export of the entries. The editor does no spell check and suggests nothing while you type.
“Delete all data” in the danger zone deletes your audio and transcription data; your dictionary, however, stays.
Terms that pay off for your dictionary
Enter what is named again and again in your recordings. This includes place names, authorities, parts of file numbers, and the technical vocabulary of your field, including abbreviations. Names of participants that show up in several recordings, or again and again, can also be useful for the dictionary (e.g. Mayer instead of Meier).
Because the space for the keyword list is limited, the selection counts more than the quantity. Ten terms that come up in every recording bring more advantages than one hundred terms from one recording.
Tip: If an error comes up again and again in one specific recording only, search and replace is the faster way. Only what applies beyond the individual case should go into the dictionary.
Where to go next
- Search and replace cleans up an existing transcript.
- Correcting transcripts gives you an overview of the editor.
- Start a transcription explains how you submit a new job.