Creating NER annotations is a two-step process: An Aviary organizational user requests NER for a transcript. An Organization Owner approves the pending request to start processing. NER analyzes text, so a resource must already have a transcript before NER can be requested. NER does not create a transcript. Any valid transcript in Aviary can be processed, whether it was uploaded, created in the transcript editor, or generated by an automated transcription service. Requesting NER annotations 1. Navigate to the Resource Detail page for the resource and open the Transcript tab. Click the vertical ellipsis (⋮) in the Transcript menu. 2. Click Request NLP - spaCy in the dropdown menu. 3. The Create Named-Entity Recognition modal window will open. Check Find associated Wikidata records to link the entities identified by spaCy to Wikidata records and extract additional metadata from Wikidata and Wikipedia. Leave the option unchecked to identify entities without linking them. Click the Create NER button to submit the request. 4. A confirmation message will be displayed in the top right corner of the Resource Detail page after the request is submitted. Approving NER requests An Organization Owner approves pending NER requests from the Automated Metadata page, which is opened from the Automated Metadata tab in the staff menu. 1. The Automated Metadata table will display a new row for the request. The service is listed as SpaCy and the status is listed as Pending. Click the Approve button to approve the request. 2. A confirmation window will open. Confirm the approval to start NER processing. 3. After the request is approved, spaCy processes the transcript. Processing time varies with the length of the transcript and the existing queue of jobs. When processing is finished, the request status will be updated in the Automated Metadata table. Note: Any organizational user can request NER annotations, but only an Organization Owner can approve a pending request. See the Approving automated metadata requests page in Automated Metadata for more information about the approval workflow.

When processing is complete, the NER annotations are displayed in the Transcript tab on the Resource Detail page. Each NER request creates a separate annotation set, numbered sequentially in the order the requests were made (for example, SpaCy NER 46). Each identified entity is highlighted in the transcript with a color that corresponds to its entity type (person, place, organization, event, or date). Click a highlighted entity to display its details in a window directly below the media player.

Plan to review every NER annotation set before sharing it with researchers. spaCy identifies entities from patterns in the text and will misclassify some of them. For example, the name “Jim Beam” may refer to a person or to the business that bears his name, and the service may assign the wrong entity type. Wikidata linking can also connect an entity to the wrong record when several records share a name. For example, a date entity “today” could incorrectly link to the Wikidata entry for the “Today” television show. NER annotations are edited in the same transcript editor used to edit transcript text. Opening the Entity Recognition tab 1. In the Transcript tab, click the pencil icon to open the transcript editor. 2. Click the Entity Recognition tab in the transcript editor. 3. Click Show Entities to highlight the entities in the transcript. 4. A key of entity types is displayed at the top of the screen. Click the eye icon beside an entity type to show or hide that type in the transcript. Viewing entity details 1. Click any highlighted entity in the transcript to display a box with basic information about it: the entity type, entity name, linked Wikipedia entry, and a short Wikipedia description. 2. Click the View Entity button to see the complete annotation. 3. The Entity Detail view will open to the right of the transcript editor. Depending on the entity type and the linked Wikidata record, the Entity Detail view displays these fields: Entity Name and Entity Type Description NEL Authority, NEL Authority ID, and NEL Authority Link: the linked authority (Wikidata), the record’s identifier (QID), and a link to the record Wikipedia Link and Thumbnail URL Geolocation coordinates (places) Organization official website (organizations) Date of Birth/Death (people) and Date (dates and events) Correcting an entity’s Wikidata link The following example corrects an entity that was linked to the wrong Wikidata record. 1. Click Entity Editing Mode in the Entity Recognition tab on the left side of the transcript editor. Entities and entity metadata can only be changed while Entity Editing Mode is active. 2. Find an entity that was identified incorrectly or was not linked to Wikidata, click it, and click View Entity. In this example, the text “Lewis” is selected. 3. Review the Entity Detail view. In this example, “Lewis” was correctly tagged as a person but was linked to the Wikidata record for Lewis, a British television series. The record is incorrect: earlier in the same sentence, the speaker refers to Anthony Lewis, whose name was correctly identified and linked. 4. If the correct entity already exists in Wikidata, click the value beside NEL Authority ID and replace it with the correct Wikidata QID. In this example, the incorrect QID Q2523064 is replaced with Q573532, the Wikidata record for Anthony Lewis. 5. Click the refresh button beside the NEL Authority ID to update the metadata extracted from Wikidata. 6. The Entity Detail view will update with metadata from the new Wikidata record. In this example, the Entity Name changes to Anthony Lewis, and the occurrence count changes from 1/1 to 1/2 because the transcript now contains two matching occurrences. Matching occurrences share both the same Entity Name and the same Entity Type. Use the left and right arrows to move between occurrences. Editing all matching occurrences To apply the same change to every matching occurrence of an entity, turn on the Modify all occurrences toggle in the Entity Detail view before making the change. A separate success message is displayed for each occurrence that is updated. Changing an entity type If spaCy assigned the wrong entity type, such as tagging an organization as a person, select the correct type from the entity type dropdown menu at the top right of the Entity Detail view while Entity Editing Mode is active. Removing an entity To remove a single entity annotation that should not have been identified, click the trash icon (🗑) in the Entity Detail view. To remove every occurrence of an entity at once, click the Delete all occurrences button below the Modify all occurrences toggle. A confirmation window will display the number of occurrences that will be deleted. Click Delete to remove them, or Cancel to keep them. Removing all entities from a transcript If a review shows that an NER annotation set is not usable, all of its entities can be removed so the transcript can be annotated again from scratch. The Remove All Entities button is only available while Entity Editing Mode is active. Click it in the Entity Recognition tab. A confirmation window will open. Click Delete to remove every entity from the selected annotation set, or Cancel to keep them. Note: Deleting all occurrences of an entity and removing all entities from a transcript cannot be undone. Adjusting the transcript editor layout Click the << or >> buttons to resize the sections of the transcript editor. Minimizing the left column opens the media file in a separate picture-in-picture player, which provides more space for reviewing transcript content and entities. Using the NER Summary table The NER Summary table in the Entity Detail section lists every entity in the transcript in tabular form, which helps reviewers assess an annotation set as a whole and apply corrections consistently. 1. Click the ↑ button in a column header to sort the table by number of occurrences, Text/Entity Name, or tag type. 2. Sorting by Text/Entity Name groups similar names together, which makes it easier to find entities that should be treated as the same entity. In this example, “African National Congress” and “ANC” refer to the same organization. Click an entity name in the table to jump to its first occurrence in the transcript editor and display it in the Entity Detail view.