Skip to Content
Pangur docs

Explore a corpus

Once a corpus has been analysed, its views show the people, places, dates and events the readings found. This page is the reference for each view, its filters, and what each one counts and does not establish.

Once a corpus's sources have been analysed, its views show the people, places, dates and events the readings found, across every source in the corpus. This page describes each view and its controls. Looking at a view reads the stored analysis and does not run a model, so it costs nothing, except where this page says otherwise. Building a corpus and analysing it are covered in Work with corpora.

The views

The views are tabs on the corpus page, and Index opens by default.

ViewWhat it shows
RecordsThe corpus's records (letters, entries, sermons, events) in date order.
IndexEntities by kind, or the words and phrases you choose to compare.
ConceptsWhat the corpus's texts turn on, from several sources of evidence.
Dates and placesA map of the places in the records, which can play through time.
TimelineDated records and events, with uncertain intervals shown as ranges.
NetworkWho writes or speaks to whom, stated relations, and who is named together.
OverviewActivity over time and the subjects the corpus holds.
TrendsCounts of records or attestations by day, month or year.
SourcesThe corpus's sources, their reading and review status, and the reading guidance. See Work with corpora.
ReviewDecisions about which names are one entity. See Review a reading.

The investigation frame

The views from Records to Trends share one frame, headed Corpus investigation, so that a filter you set in one view applies in all the others.

Research question chooses what kind of evidence the views gather. The choices are Correspondence routes (who wrote to whom, from where to where, offered only when the corpus has letters), People's whereabouts, Movement evidence, Mentions and activities, and Words and phrases. A note beside it says what the dates in the views mean for the question you chose: the dates records were written, the dates of the occurrences for whereabouts and movement, or the dates of the events described.

Dates and filters opens the filters described in the next section. The frame reports how many records match (for example "212 of 640 records") and lists the active filters, each with a way to remove it, or Clear filters to remove them all. Focus narrows every view to one entity, such as "Focus: John Keith", and the × clears it. When an entity is in focus, Subject's role decides which records count for it: all evidence, the records it wrote or spoke, the records it wrote, spoke or was addressed in, or the records that mention it.

View as results shows a view's content as a list that can be used with the keyboard, and View visualisation returns to the chart or map.

Filters

FilterWhat it matches
Writer or speakerWho wrote or spoke the record.
Addressed toWhom the record was addressed to.
Written or spoken atThe place the record was written or spoken.
Sent toThe destination of a letter.
DatesFrom date and Through date, or A single year (for example 1709). Dates can be years or full dates. The start date must precede the end date.
SourceOne source in the corpus, or All sources.
Include undatedWhether records with no date are counted.

Each name filter has its own search box. Note that the filters apply to what the readings found, so a record that the reading did not attribute to a writer is not matched by Writer or speaker.

Index

The index lists the entities found in the corpus by kind: People, Places, Organisations, Works, Events, Concepts, Objects and Other subjects. Choosing an entity opens its evidence. To compare expressions instead of entities, choose Words and phrases under Index of.

Words and phrases

Words and phrases is a literal concordance. Pangur counts the exact expressions you give it, without a model. Under Edit expressions, write each group as a label followed by its variants, for example light: inward light | light within | inner light. You can compare up to 12 groups of up to 30 variants each. Matching is case-insensitive, and there is no automatic stemming and no synonym list, so add each form you want counted.

The results have three tabs. Usage gives each group's rate per 10,000 words, and Compare by splits it by period, author or source. Passages lists the matching passages, sortable by date, author or source. Associated vocabulary lists words that occur near the expressions.

Editorial matter, such as an editor's summary, is left out of the counts where the source marks it as editorial.

Records

Records lists the corpus's records in date order with their writer, recipient and date. Find a record or person… searches them, and the year list (All years, a single year, or Undated) narrows them. Choosing a record opens its evidence, including its text, entities, events and dating.

Timeline

The timeline places dated records and events in time. A date the source gives only roughly, such as a month or a range, is drawn as an interval. What is drawn depends on the research question chosen in the frame. If nothing in the current filters is independently dated, the timeline says so rather than drawing anything. Events whose quoted words were not found in the text are left off.

Dates and places

The map shows the places the records name, using coordinates from the authorities they are linked to. A place without coordinates is not drawn, which is one reason to link entities to authorities.

Place roles chooses which places are drawn: origins, destinations, mentioned places or event locations. Play through time animates the map by day, month or year, either showing each period on its own (Change within a period) or adding to what came before (Cumulative).

The map is careful to state what its marks do not show. For letters, the arrows join the recorded origin and destination, not the postal itinerary, and the dates are when the letters were written. For mentioned places, the map notes that a place mentioned in a record does not establish the subject's physical presence there. Lines that follow a person join consecutive dated observations; the routes and dates of travel between them are unknown.

Network

The network draws entities as nodes and their connections as lines. Connections chooses which connections are drawn. The default, who writes or speaks to whom, comes from the writer and addressee of each record and is drawn as solid arrows. Stated relations are the relations the readings found, such as kinship, drawn as ochre arrows. Named together joins entities named or present in the same record, drawn as dashed lines.

Entity kind limits the nodes to people, places or another kind, and Find an entity… finds a node. A large network is cut down for display, and a note says how much is shown ("Showing 200 entities and 900 connections. Narrow the filters for the rest."). For keyboard access, use View as results.

Overview

The overview summarises what the corpus holds. Activity over time charts observations, events and records by date, and Subjects counts the entities of each kind and opens the index at that kind.

Trends counts the filtered corpus over time. You choose the interval (day, month or year), what to compare (person, place, activity or source), and whether to count distinct records or attestations. The counts are of what the corpus holds, and the view says so: they are corpus counts, not population rates.

Concepts

The Concepts view lists what the corpus's texts are about. Each concept says where it comes from:

OriginHow it is found
Named in the recordsConcepts the readings found named in the text.
A topic the analysis gaveTopics the readings gave to units.
A group the records form by meaningRecords grouped by a local text-similarity model on Pangur's server. A model then names a group only where quotations from it support the name.
A theme found in reading every recordThemes from the corpus digest.
Your conceptA concept you asked Pangur to test.

Groups and themes appear a few minutes after the corpus changes, once Pangur has read it in the background. Until then the view says that there are no concepts yet.

Testing a concept

You can name an idea you expect to find in the records and have Pangur test it. Under Test a concept, give a Name (for example "the inner light"), What it is, in a sentence, and optionally Words that may voice it, separated by commas, then choose Test this concept.

Pangur first counts the concept's words in the records. It then has a model judge the passages most likely to contain the concept, together with other passages chosen for comparison, without telling the model which passages you expect to match. The verdict is one of Borne out, Weakly borne out, Not found or No name fits, shown with the counts it rests on, for example "Found in 9 of the 12 closest records and 1 of 12 others read for comparison." While the test runs, the concept is marked as being tested.

A tested concept is never adopted on its own. You record your decision with Keep or Set aside, and Clear decision undoes it. If the corpus changes after the test, the concept is marked as tested on an earlier state of the corpus.

The model judgement in a concept test is charged to the project owner's model allowance. If the test cannot start, the concept is counted by its words only, and Pangur says that the full test could not start now.

Evidence

Choosing an entity, record or event in any view opens its evidence beside the view. For an entity, the evidence lists each mention with its quotation, page and source, and Only this corpus or Show all switches between this corpus and every source in the project. A quotation that was not found in the document is marked as unsupported. Read the records that name this opens those records in the reader. The evidence also carries the entity's authority link and its merge history, both described in Review a reading.

Source reading

Corpora whose sources were described by a source inventory have a source reader for each record. It shows the pinned source text with its parts (Letter text, Enclosure, Supplied translation, Editorial summary, Editorial commentary), the Evidence revision and Historical reading it rests on, its Attestations, and a Photograph of the page where one has been linked.

On this page