1. Introduction
Plaid IGT is an editor for interlinear glossed text (IGT), for documentary and field linguists. It runs in a web browser and covers transcription, interlinear analysis, and the lexicon in one program.
Key features include:
-
Transcription. Attach audio or video to a document and transcribe it one time-aligned segment at a time, with playback speed control and a speaker label on each segment. The transcript is the text that is then analyzed.
-
Interlinear analysis. Divide the text into sentences, words, and morphemes, and annotate them with the fields your project defines: glosses, parts of speech, free translations, and others.
-
Vocabulary linking. Words and morphemes may be linked to entries in a vocabulary. A linked entry supplies its gloss wherever the form occurs, so a word is analyzed consistently across the corpus. Homonyms, affixes and clitics, and multi-word expressions are supported.
-
Import and export. Import a FieldWorks (FLEx) backup or interlinear texts, ELAN files, or a CLDF dataset. Export to FLEx (
.flextextwith a LIFT lexicon), ELAN (.eaf), CLDF, plain text, or a LaTeX book, or copy a sentence as LaTeX. -
Collaboration. A project is shared with its members, each of which may have reader, writer, or maintainer privileges. A central database lets all users see the latest state of the data, with no import/export required.
-
Audited edits. Every change is recorded in an audit log with its author, and any document can be viewed as it was at an earlier time.
-
Pluggable AI assistance. A connected AI model can transcribe a recording, draft translations, or propose segmentations and glosses. Machine output is marked until a person confirms it, and human work is never overwritten.
-
Special character entry support. Praat’s backslash codes work in every field:
\swfor ə,\ngfor ŋ,\?gfor ʔ. Additional codes may also be defined. -
Comments. Any sentence, word, morpheme, or annotation can carry a comment thread to allow in-context discussion of analyses.
-
Bulk edit and query tools. Search every document by form or annotation, respell a word across the corpus, and hold a field to a controlled list of tags.
2. Getting Started
Plaid IGT is served by your Plaid server. There is nothing to install.
Open the editor in your browser at the /igt/ path on that server.
If you are running Plaid locally with the default settings, that is:
http://localhost:8080/igt/
Sign in with the email address and password your project administrator gave you, or with the password you chose when you opened an invitation link (see Sharing and Permissions). Your name at the top right opens a menu with Profile and Sign out. On your profile you set a display name and a picture, change your password, and mint API tokens.
The editor has a few levels you move between:
-
Projects: the list of all projects you can access. This is your home screen after signing in.
-
Documents: the texts inside one project. Beside the document list sit the project’s own tabs: Search, Guidelines, Assistant (when an assistant is running), and Export for everyone, and Bulk Edit, Validation, Activity, and Settings for maintainers.
-
A document’s tabs: Baseline, Media, Tokenize, Analyze, Comments, Export, and Details, switched at the top of an open document.
The Vocabularies link in the header leads to the vocabularies (see Vocabularies), and Guide opens this document.
|
Tip
|
The Tokenize, Analyze, and Media tabs each have a ? button that opens a legend of the gestures and keys that tab answers to. If you forget a shortcut, that is the quickest place to look. |
3. Projects
A project holds a collection of documents along with everything they share: the annotation fields, the orthographies, the tagsets, the vocabularies it draws on, and its members. Create a project from the Projects screen, either blank (a short wizard asks for those settings) or from imported data (see Import and Export). Click a project to open its documents.
Because so much is configured per project, two projects can look quite different. A project for a field methods course might have a single Gloss field and a free translation, while a documentation project adds a part of speech, a phonetic orthography, and a lexicon.
3.1. Sentences, words, and morphemes
Interlinear glossed text is organized at three scopes, and the annotation fields and the Analyze grid are built around them:
| Scope | What it is |
|---|---|
Sentence |
A stretch of the baseline text. Sentence fields hold things like the free translation, the speaker, and notes. |
Word |
A token within a sentence. Word fields hold things like a part of speech or a word-level gloss. |
Morpheme |
A piece of a word. The morpheme gloss and any other morpheme fields live here. |
A word is divided into one or more morphemes. A simple word is a single morpheme, while a word like English cats is two, cat and -s. Every field in a project belongs to exactly one scope, which decides which row of the grid it appears on.
3.2. Documents
The Documents tab lists a project’s texts with their word counts, when each was last changed by anyone, and when you last changed it yourself. Click a column heading to sort by it, and again to reverse it. Sorting by Your last edit is the quickest way back to what you were working on. New document offers Blank document, which asks for a name and opens the new, empty document on its Baseline tab, and Import from ELAN (see Importing). A member who is not a maintainer is asked for the name straight away.
Each document is built up on its tabs, which the rest of this chapter walks through.
3.3. Details
The Details tab holds the document’s name and its document-level fields, such as speaker, date, and notes. Which fields appear is configured per project (see Annotation), and a field can be held to a tagset, so that a genre is always one of a fixed set (see Tagsets). This is also where a document is renamed, copied, and deleted. Copy makes a second document in the same project, with the same text, annotations, vocabulary links, and recording, but no comments.
Text direction says which way the document’s words are laid out. On Automatic, a document written in Arabic, Hebrew, or any other right-to-left script lays its words out right to left in the Analyze grid, with the row labels on the right. Set it by hand for a document the text cannot speak for, such as a right-to-left language written in a Latin transliteration. Each field still reads in its own direction, so an English gloss reads left to right under an Arabic word.
3.4. Baseline
The Baseline tab holds the raw text you are analyzing, in whatever orthography you work in. Paste or type it, one sentence per line, and save. On the first save each line becomes a sentence. Each save also splits the text it adds into words, unless the project has Tokenize new text off (see Annotation). A script written without spaces between words, such as Chinese, Japanese, or Thai, is not split into words, so make its words on the Tokenize tab.
You can edit the baseline later too, and the sentences, words, and annotations already there are kept. A word you change keeps its analysis as long as one of its letters is left. A word all of whose letters you replace, by selecting it and typing or by pasting over it, is deleted with its annotations, even when you type the same word again, and what you typed is new text. When you select two or more words and type or paste over them, letters you type again at either end are kept only as whole words: select cat sat and paste c, and both words go. A word whose text you delete is deleted with its annotations too. A sentence whose exact text you replace keeps its translation and its other fields, and a segment its times. A paste or a retype that covers a whole sentence or segment and more text deletes it, with its translation or its times. A save that deletes words, segments, or sentences with something on them asks first, with counts. New text becomes words of its own, and text typed at either end of a sentence joins that sentence. A new line typed after a sentence joins that sentence, and one typed before the first sentence becomes a sentence of its own. To make a new line a sentence, Ctrl/Cmd+click its first word on the Tokenize tab. A space typed inside a word splits it in two. The part holding more of the word keeps its analysis, and the other part is a new word with none. Deleting a space does not merge words. Do that on the Tokenize tab. Ctrl/Cmd+Z in the box undoes typing, and Ctrl/Cmd+Shift+Z redoes it.
When someone else saved the baseline while you were editing it, Save changes puts your changes onto their text. If you both changed the same passage, the save is refused and your text stays in the box. Cancel then shows the saved text, and you make the edit again.
A project may define additional orthographies, alternate representations of each word such as a phonetic transcription or a practical spelling. These appear as their own rows in the Analyze grid (see Text and vocab).
3.5. Media
The Media tab attaches an audio or video recording to the document and holds its transcript as time-aligned segments. Each segment is a stretch of the recording with its text, which becomes part of the baseline, and an optional speaker label. Media is optional, and a document with none is simply a text. Common audio and video formats are accepted (MP3, WAV, M4A, MP4, WebM, and others). Until the whole file has gone up, Cancel stops the upload, or the conversion or mix before it, and nothing is stored.
A recording has to be one the browser can play.
A WAV that browsers cannot play, such as the ADPCM files ELAN and some recorders write, is converted to plain PCM when you choose it.
Any other file the browser cannot play is refused, as in story.wav cannot be played in a browser (GSM 6.10 WAV).
Convert such a file to MP3 or PCM WAV with an audio editor first.
A large recording can be sent as audio only. Choosing one under Single offers to convert it to MP3 in the browser, mono at 16 kHz, about 14 MB an hour. The timing is unchanged, but the picture is dropped and the sound is no longer good enough for phonetic measurement.
3.5.1. Recordings with one file per speaker
The upload card has two modes, Single and Per speaker. Single takes one file, and the recording plays as it was recorded, a stereo one in stereo. Per speaker is for a conversation recorded with a microphone per speaker.
A microphone per speaker gives several files that share one clock. Under Per speaker, choose them all, and Mix recordings makes one recording of them with a channel per file. Each row takes the speaker of that file and an offset in seconds, silence put before the file (a negative offset trims its start) for a file that was started late. A stereo file is averaged to one channel first. Full quality keeps the highest sample rate among the files, Smaller resamples to 16 kHz, and each says the size it comes to. The dialog starts on Full quality, or on Smaller when only that one fits the server’s limit. A mix the server would refuse is refused here, before anything is uploaded. A file the browser cannot decode is marked Cannot be decoded., and its Remove takes it out of the mix.
A file made elsewhere with a speaker on each channel can be chosen alone under Per speaker. It is not mixed. Speakers per channel takes a speaker for each channel. A WAV is uploaded as it is. Any other audio file is uploaded as a WAV with the same channels, at Full quality or Smaller as a mix is. A video file chosen alone is refused, a WebM or MP4 that holds only sound included: upload it under Single.
A recording uploaded under Per speaker shows a channel list under the player. Each channel has its speaker, which you can change, and Mute and Solo, which only change what you hear. Remove speakers per channel under the list makes it a plain recording again. Speakers go on channels only at upload, so to have them back the recording is uploaded again.
With the speakers filled in, the timeline draws one waveform lane per channel in the speaker’s colour, speech detection and transcription list the channels so each runs on one channel alone, and every proposal and segment they make carries that channel’s speaker. Two speakers' segments may overlap in time.
The mix is made once, when the files are uploaded. To add a speaker later, delete the recording and upload all the files again.
3.5.2. Transcript
The transcript is a list of segments in time order, and it is built for typing what you hear. Play the recording, then type the utterance you just heard into the New segment row and press Enter. The new segment runs from the end of the previous one to wherever playback is when you save. It takes the speaker of the last segment made in this document since the page was loaded, so a run of turns by one speaker needs nothing more than Enter.
Every segment is a row you can edit:
-
Moving into a row plays its segment. Turn off Play segment on entry to play by hand instead.
-
Enter saves the row and moves to the next. Tab moves to the next field within a row.
-
↑ and ↓ move to the row above or below, from the start or end of the text.
-
Esc puts a row back the way it was.
-
Shift+Space pauses and resumes the segment you are in without leaving the row.
-
Shift+← and Shift+→ move playback one second.
-
Space plays and pauses the whole recording when you are not typing in a box.
-
Alt+Enter splits the segment at playback, its text between the words at the caret, and Alt+M merges it with the next segment. The words, their annotations and the sentences are unchanged, so two segments in different sentences do not merge.
A change to a row’s text is saved as it is on the Baseline tab, and the words in it keep their analysis unless all of a word’s letters are replaced. A segment whose exact text you replace keeps its times. When someone else changed the same segment’s text while you were editing it, the row shows their text with yours under it. Enter keeps yours, and Esc lets it go.
A segment’s start and end times are boxes of digits in its row. Type into the box under the caret, or step it with ↑ and ↓: milliseconds by 10, or by 100 with Shift. A segment can never run into its neighbours. The speed slider runs from 0.25× to 5×, and the loop button repeats the segment you are on until you pause.
The trash on a row deletes the segment. When its text carries annotations, it asks first: Delete text deletes the text and its annotations along with the segment, and Keep text leaves the text in the baseline. Delete segments, in the Transcript header, removes segments in bulk: all of them, the ones with no text, or the ones whose text is a value you give.
Segments typed in the transcript go onto one line of the baseline. Split at segments on the Tokenize tab makes each of them a sentence, and the transcript links to it.
Transcribe, also in the Transcript header, has a transcription service write segments and their text into the document (see Services). Its segments are marked as machine-made until you edit or confirm them. Everything already in the document stays, so running it again adds only what is new. To transcribe from scratch, clear the text on the Baseline tab first.
3.5.3. Timeline
The timeline shows the same segments against the waveform. Drag across an empty stretch to add a segment anywhere in the recording, drag a segment’s edge to trim it, and click a segment to hear it and edit its row. Ctrl/Cmd+wheel zooms the timeline, and the wheel alone pans it. The ruler above it shows the time, and a click on it moves playback there. The card over a selected segment has buttons to split it at playback and to merge it with the next one.
A dragged stretch takes either new text or text already in the baseline. Under Existing text, select the words the segment covers. This is how you align a text that was typed before the recording was attached.
The timeline colors each speaker’s segments alike. Two segments may overlap in time only when both have a speaker and the speakers differ, which is how cross-talk is recorded. A dragged edge stops at any segment it may not overlap.
3.5.4. Speech detection
The waveform button in the Recording header detects speech in the recording and proposes where the utterances are. It runs in the browser and needs no service.
A proposal is a stretch of time with no text, dashed on the timeline and in the transcript. Type into one and it becomes an ordinary segment with those times. The × on its row discards it, and Discard in the detection dialog discards them all. Proposals are kept with the document, so they can be transcribed another day or by someone else.
The built-in detector’s options change what is proposed:
-
Speech threshold: how sure the model must be before a moment counts as speech. Raise it on a noisy recording.
-
Shortest silence: a pause shorter than this does not end a segment.
-
Shortest segment: a stretch of speech shorter than this is dropped.
-
Longest segment: a longer stretch is split at the widest pause inside it.
-
Padding: added to each end of a segment.
Detection leaves alone any stretch that already has a segment, so you can run it again after using some of what it proposed.
3.6. Tokenize
The Tokenize tab divides the baseline into sentences and words. Words are highlighted, and text that is not yet part of any word is plain. Nothing is annotated here. That happens on the Analyze tab.
To tokenize by hand:
-
Drag across plain text to make a word of it, or across several words to merge them.
-
Click a word to split it at a character. Esc closes the split points.
-
Right-click a word to delete it.
-
Ctrl/Cmd+click a word to split its sentence there, making that word the first of a new one.
-
Alt+click a word to open it on the Analyze tab.
-
The ↑ button on a sentence merges it with the previous one.
A change that would delete annotations asks first, and counts them.
To tokenize the whole document at once, use Tokenize in the Tokens header. The built-in Rule-based punctuation splitter finds words and leaves sentence breaks as they are, and a tokenization service can split sentences too (see Services). Words already there are kept, so text added later can be tokenized by running it again. Clear tokens and Reset sentences, beside it, start over. Split at segments adds a sentence break wherever a segment from the Media tab starts, so a recording cut into utterances becomes one sentence per utterance.
A project can mark certain tokens as ignored, usually punctuation.
Ignored tokens are not words. They read as plain text everywhere, in the grid, in counts, searches and exports, with no annotation and no morphemes.
The built-in splitter breaks words at punctuation, so an orthography that writes a glottal stop or an ejective with an apostrophe needs that character listed under Characters that behave as letters in Settings → Annotation (see Annotation).
Then k’a stays one word.
|
Tip
|
A word that came out split can be repaired by dragging across its pieces, which merges them into one word. |
3.7. Analyze
The Analyze tab is where interlinearization happens. It shows each sentence with its words, the morphemes under each word, and a row for every annotation field. Long documents are shown twenty-five sentences at a time, with a pager above and below the grid. A document of more than 15,000 words is marked Long document, since editing slows down past that length. Split it into shorter ones.
What you do here:
-
Segment morphemes. Break each word into morphemes and edit their forms. Type
-in a morpheme to split it at the cursor, or=to split at a clitic boundary (see Morpheme types). Backspace at the start of a morpheme merges it into the previous one. Alt+0 types the zero morph ∅ (see Typing Special Characters). -
Gloss and annotate. Fill the gloss and any other word- or morpheme-scope fields your project defines. The editor offers faint guesses, such as a gloss it has seen before for the same form, which you accept with Enter or type over.
-
Pick from what the project already uses. On a cell, Alt+↓ lists every value the project has given that form, with counts. On a morpheme field with a lexicon field, such as Gloss, each row pairs a value with the lexicon sense it goes with, and taking one sets both. A field held to a tagset opens its list on its own and narrows it as you type.
-
Respell a word. Respell in a word’s popover, or F2 on any of its cells, opens a box over the word with its text selected. Enter saves the new spelling, and Esc or clicking elsewhere cancels. The word keeps its morphemes, glosses, lexicon links, multi-word expressions, and comments. A word with one morpheme spelled as the word gets the new spelling there too, and a word split into morphemes keeps their forms as they are. A spelling must have at least one character and no space: split a word on the Tokenize tab, and make larger changes on the Baseline tab.
-
Link vocabulary. Connect a word or morpheme to an entry in a lexicon the project links (see Vocabularies). When the same form occurs unlinked elsewhere in the text, all ×N in the popover links every one of them at once.
-
Reuse an analysis. Once you have analyzed a word, its popover offers Analyze every "…" in this text like this, which copies the analysis onto the unanalyzed words spelled the same. To draw on the rest of the project, use Auto-analyze (see Copying previous analyses).
-
Auto-analyze. Let the machine make a first pass over the whole document: translations, analyses copied from earlier work, a model’s segmentation and glosses, and lexicon links (see Auto-analyze).
-
Review machine work. Ctrl/Cmd+Enter accepts everything proposed on a word and moves to the next. Ctrl/Cmd+Backspace discards a word’s proposal. Ctrl/Cmd+Shift+↑/↓ jumps between the words that still have one.
-
Fill sentence fields. Record the free translation, the speaker, and other sentence-scope values on each sentence’s row.
-
Show only the rows you need. Click a row’s label to choose which rows show. A minimized row stays as a thin stripe you can click back open.
-
Copy as IGT. Copy one sentence for a paper or a message, as aligned plain text, tab-separated text, LaTeX (gb4e or ExPex), or leipzig.js HTML. The LaTeX keeps characters like ə and ∅ as they are, so typeset it with XeLaTeX or LuaLaTeX. In LaTeX, gloss parts written in capitals, such as NOM or 1SG, are set in small caps.
Move around the grid with Enter or Tab (next cell, Shift for the previous one), ↑ and ↓ (rows), and ← and → from the ends of a value. Alt+click a word to open it on the Tokenize tab. The full set of gestures is collected under Keyboard Shortcuts.
3.7.1. Morpheme types
A morpheme’s type, such as stem, suffix, or enclitic, decides how it joins its neighbours: = beside a clitic, - otherwise.
In the grid, a proclitic’s or an enclitic’s = sits against the clitic, with a narrow space toward its host: ni= house, house =ni. Any other = sits in the middle.
Copied and exported text writes the plain =.
Typing = in a morpheme form splits it and types the clitic by its place in the word.
A = after the word’s first morpheme makes that morpheme a proclitic, and a = before the last makes the last an enclitic. In a word of two morphemes, the second is the enclitic.
A = between two inner morphemes types neither, and still shows as =, in the grid and in every copy and export.
A FLEx export writes such a = in front of the morpheme after it, as FLEx writes an enclitic, and a proclitic’s = once, on the proclitic. Before a morpheme typed as anything but a clitic, such as a stem or a suffix, it writes none, since FLEx has no way to say it. An ELAN export writes every = in front of the morpheme after it.
Each = typed is kept with the morpheme after it, and the types are worked out again over the whole word each time its morphemes are split, merged, or deleted, so the result does not depend on the order they were typed in.
Merging the two morphemes beside a =, or deleting the morpheme after it, takes the = away, and so does deleting a word’s first morpheme for the = after it.
Clearing a clitic type in the Type list takes away the = that gave it. Clearing any other type keeps the =, and a = the cleared type blocked then types the morpheme at once.
A type chosen in the Type list, given by an import, or carried by a linked entry is never changed this way.
A copied analysis (Analyze every, Auto-analyze, Bulk Edit) carries each = with it, and a CLDF import types the pieces of a word written with = by the same rule.
The Type list in a morpheme’s link popover sets its type, and — clears it. On an unlinked morpheme it sets the morpheme’s own type. On a morpheme linked to an entry with a type, it shows the entry’s type, and a change applies to every use of the entry. Only writers and maintainers of the vocabulary’s project can change it. Linked to an entry with no type, the morpheme goes by its own, and the list shows and sets that, not the entry’s. While the form of the + row is open, the list reads Type (new entry) and sets the type of the entry being created. It starts on the type the list showed before (the linked entry’s, else the morpheme’s own), or on — when there is none, and an entry created as is takes that type. A form typed with a marker, such as -s, sets the list to the type the marker names, unless a type was picked from the list. Closing the create form with Esc brings back the list of the linked entry or the morpheme. In a project with no lexicon, the set type button under a morpheme opens the list.
3.7.2. Provenance
Plaid IGT tracks who, or what, produced every value. Four states are distinguished in the grid:
-
A value you typed shows plain. A marked value you have edited or confirmed keeps only a thin violet underline.
-
A machine-made value, such as a gloss a service proposed or a vocabulary link Auto-analyze made, shows in violet italic. It is in the document, and it waits for a person to accept or discard it. When the service that made it gave its confidence, hovering shows the percentage, and a value below the project’s Low confidence threshold gets a heavier, darker underline. A maintainer sets the threshold on the project’s Services settings page.
-
A contributed value shows in amber italic. It was entered by someone whose work the project reviews (see Access), and it waits for a verifier to accept or discard it.
-
A guess is a faint gray placeholder in an empty cell. It is not in the document at all until you take it with Enter. A guess on a pale teal background comes from the linked lexicon entry, and one on plain background from what the same form was given elsewhere in the project.
Hover over an annotation to see where it came from. Editing a marked value confirms it, so the violet and amber italics disappear as you work through a document. The small form beneath a word or morpheme is the lexicon entry it is linked to.
A member whose work is reviewed is a contributor. Everything they enter is marked contributed, including a machine proposal they accept. Everyone else is a verifier, whose work shows plain and whose edits and confirmations settle marked values.
A value that a field’s tagset refuses is underlined in red, and a closed tagset will not let it be entered at all (see Tagsets).
3.7.3. Saving
A value is saved when you leave its cell, and there is no save button. When the server cannot be reached, the save status reads Can’t reach the server, retrying, or Offline, retrying when the computer has no network. Your edits are sent once the server answers again. Closing the tab meanwhile asks first. On every tab but Analyze the save status is beside the tabs, and a save the server has not answered after three seconds reads Saving… there until it lands.
Two people can have one document open at once, but neither sees the other’s edits as they are made.
Edits to different fields are all saved.
An edit to a field that someone else changed anywhere in the document since you opened it can be refused with Changed elsewhere, and your value stays in its cell, not saved.
When someone else changed the same cell first, the cell shows their value with an amber edge and yours under it, and a message names the change, such as b changed this to DOG.
Enter keeps your value, and Esc or typing another value lets it go.
Any other edit that clashes with a change elsewhere is refused, and the document shows the latest version for you to redo it.
3.8. Comments
Any sentence, word, morpheme, or annotation value can carry comments: a conversation in the margin of the text, for a question to the speaker, a doubt about a gloss, or a note to a classmate. In the Analyze tab, hover a sentence’s label, a word, or a morpheme and click its comment badge to read the thread or add to it. A badge with a number marks a thread that already has comments.
The Comments tab lists every thread in the document, newest activity first. Click a thread to open it, and click its arrow to jump to its sentence. Comments take Markdown. You can edit your own, and a maintainer can remove any.
A comment stays even when what it was about is edited away, such as a word that is merged or retyped. It is then listed under the tab’s Outdated filter until someone deletes it. Comments are not part of a document’s history (see History), and are left out of every export but the native archive.
An entry in a vocabulary takes comments too. Open the entry to read or add to its thread, and the vocabulary’s own Comments tab lists every thread.
4. Vocabularies
A vocabulary is a lexicon: a list of entries (words or morphemes) that carry their own fields, such as a gloss or a part of speech. Every vocabulary belongs to one project, and that project’s roles decide who may change it (see A vocabulary’s project). Other projects can link it too, and there it is read only to anyone who is not a member of its own project. While you analyze, you connect a word or morpheme to an entry, by hand with the link button on the word or morpheme, or in bulk with Auto-analyze.
Linking keeps the same word analyzed the same way everywhere it appears. A linked entry offers a guess in each field that has a lexicon field (see Annotation): the entry’s value in that lexicon field. Editing an entry changes it for every document linked to it, in every project that links the vocabulary. A value already in a document’s field stays as it was.
In the link popover, the linked entry’s form opens the entry in its vocabulary. When no entry fits, the ** row under the list adds one with the word's or morpheme's form and links it. The new entry also takes the values the word or morpheme holds in fields that have a lexicon field, such as its gloss. On a word or multi-word expression that no entry is spelled like or starts with, the popover starts on the ** row. Writers and maintainers of the vocabulary’s project can add an entry this way, in any project that links it, and nobody else is offered the + row.
An entry can carry a morpheme type: stem, prefix, suffix, enclitic, and so on, following the FieldWorks inventory.
A linked morpheme takes its type from the entry, or goes by its own when the entry has none. The type decides how it attaches to its neighbours in the grid and in exports (= for clitics, - otherwise).
In the Analyze grid, a morpheme’s link popover sets the type (see Morpheme types).
Writers and maintainers of the vocabulary’s project can change an entry’s type.
Vocabularies have their own area in the app, under Vocabularies in the header, grouped by the project each belongs to. There you browse and edit entries and define the fields an entry carries. Which vocabularies a project draws on is part of the project’s settings (see Text and vocab).
4.1. A vocabulary’s project
A vocabulary belongs to the project it was made in, which its page names under its title. That project’s members have the same role in the vocabulary:
-
Reader: reads the entries.
-
Writer: also adds entries and changes anything stored on one, such as its fields, type, senses and examples.
-
Maintainer: also renames, merges, deletes and restores entries, changes the vocabulary’s settings, and renames, moves or deletes the vocabulary.
New vocabulary makes one in a project you maintain, and a new project or an import makes its vocabularies inside the new project.
Another project can link the vocabulary when a maintainer of that project is also a member of the vocabulary’s project. There its members read the entries and link words and morphemes to them, and change an entry only by their role in the vocabulary’s own project. The vocabulary’s Settings tab lists them under Used by other projects, and Unlink ends one’s link and removes its links to the entries, after saying how many there are. The unlink is listed in that project’s Activity and in the vocabulary’s History. Morphemes there that took their type from an entry keep it when you can write that project, and have none otherwise. A project you cannot open is listed without its name.
A maintainer of two projects can move a vocabulary from one to the other on its page, and the old project keeps it linked. A vocabulary cannot be deleted while another project links it, and neither can its project.
4.2. Managing entries
A vocabulary’s entries are shown on its Entries tab as a table you can search, and sort by clicking a column heading.
The search box reads every column, or one field chosen beside it: pick Gloss and type go to see the entries glossed with it.
With a field chosen, a count of the entries that have nothing in that field appears under the box, and clicking it shows just those.
A Uses column counts the words and morphemes linked to each entry, and an entry’s Concordance shows every place it is linked.
Writers and maintainers of the vocabulary’s project add entries and edit their fields here. Enter in a one-line field, or Ctrl/Cmd+Enter anywhere in the entry, saves it. Only a maintainer of the vocabulary’s project renames, merges or deletes an entry.
Bulk Add takes rows pasted from a spreadsheet or read from a CSV or TSV file.
Save an .xlsx or .ods file as CSV first.
You map the columns onto the entry fields, and each row is checked against the existing entries before anything is created, so a homonym stays separate and an existing entry is not overwritten unless you say so.
When a file has a title or extra lines above its header, set the row the table starts at.
The same value on every row gives every new entry one value in a field, such as a source or a status.
Replace finds and replaces text in one field of every entry, the form included for a maintainer, as plain text, as a whole value, or as a regular expression.
Plain text ignores case, and keeps each match’s capitals when both boxes are typed in small letters: ka to ga makes Kalamang Galamang.
A fourth kind, is empty, fills the field on the entries that have no value in it.
Each entry it would change is listed with its old and new value, and only the ticked ones are written.
Export downloads the table as a tab-separated file that opens in any spreadsheet.
4.3. Entry fields
Every entry has a Form, its reference form or headword, which is required. Beyond the form, an entry carries a set of fields. A new vocabulary starts with a few built in:
-
Morph Type: whether the entry is a stem, a prefix, a suffix, and so on. The editor uses it to render affixes, such as the joining hyphens on the form and gloss lines. An entry’s form is stored without affix markers. Typed or imported as -s with no morph type, it is the suffix s. Given a morph type, as a FLEx import or the entry form gives one, it keeps that type: a bound stem written theŋjən- is the bound stem theŋjən. A form with spaces inside its markers, such as -du, du-, is kept as written.
-
Gloss: the entry’s gloss, offered as a guess in each field whose lexicon field is Gloss.
-
POS: the entry’s part of speech.
-
Definition: a fuller definition or note.
-
Status: how far along the entry is, held to a closed list (machine, draft, reviewed, published). Machine, shown as Machine draft, marks an entry an assistant job created (see Machine draft entries). The values machine and published are reserved: neither the tagset editor nor Replace renames or removes them.
You can add your own fields, reorder them, and remove any except Form, Morph Type, Gloss and Status. Every vocabulary has a Status field, and its role, its type and where it is shown cannot be changed. A field marked inline also shows on each entry’s row in the link popover. Other fields stay on the entry’s own edit form.
Each field’s role says what its values are: a gloss, a definition, a part of speech, or the entry’s status.
The FLEx, CLDF and LaTeX exports write each field by its role, and the assistant reads them.
A field with no role is none of these, whatever it is called.
A new vocabulary’s Gloss, POS, Definition and Status fields hold their roles, an import gives its fields theirs, and a field you add is given the role its name suggests, such as Gloss for Gloss (fr).
Several fields can share a role, such as a gloss in each language, and each records its language.
A field can be held to a tagset (see Tagsets), so a part of speech is always n and never sometimes noun.
A vocabulary keeps its own tagsets, under its Settings, because it is shared across projects.
A closed tagset becomes a dropdown on the entry form and refuses a value outside the list in Bulk Add.
The table counts the entries already outside it, and a click shows only those.
An entry in a vocabulary linked to a UMR project also has a UMR band, below its fields.
Roleset is the concept the entry stands for on a UMR graph, such as lunch-01. Left empty, the entry stands for its headword.
Arguments describe the roleset’s roles, one row each: a name (ARG0, ARG1, and so on) and a description (the one eating).
4.4. Headwords and senses
Every vocabulary is a dictionary. An entry is a headword, and it can have senses under it, each of which can have senses of its own. A word with one meaning is simply an entry with a gloss.
Numbers say what is what: a headword with senses is adidi 1, its senses adidi 1.1 and adidi 1.2, and a sense of the second adidi 1.2.1. Entries that share a form are told apart by that first number, so two entries spelled a are a 1 and a 2. A headword that is alone and has no senses has no number at all. On screen the number is a subscript after the form (kai₁), as FieldWorks writes a homograph number, and in exports and in the history it follows a space.
Add sense, on an open entry, writes a new sense directly under it. Set parent makes any entry a sense of another. Add headword raises a new entry over the open one, with the same form, and makes the open one its first sense, which is how a word with one meaning gets a second. A sense with no morph type of its own goes by its headword’s.
Senses, beside an open entry, unfolds the whole entry as a numbered tree. Drag a sense in it to reorder it, onto another sense to nest it there, onto Make separate entry, or onto Move to other entry. When several entries share a form, the number after the headword is a button, and dragging in its list renumbers them. The entry list opens in a By entry view that draws senses indented under their entry, beside a Flat view.
A field can hold a reference to another entry. On the Settings tab, set the field’s Type to Entry for one reference or Entries for a list, and name it for what it means: Variant of, See also, Compound of. On the entry form such a field is a picker, and the entry it points at lists the reference under Referenced by. References stay within one vocabulary.
A field can also be set to Headword only, so it is offered on a headword and not on its senses: an etymology or a pronunciation belongs to the entry, a gloss to each sense.
Examples are chosen from the texts, never typed in. Use as example, beside each row of an entry’s Concordance, adds that sentence to the entry’s Examples panel, with its translation and a link into the document. When the word is later deleted or linked elsewhere, the example says so and can be removed. A sentence worth citing that is not in any text belongs in a document, where it is glossed like any other.
Deleting an entry makes its senses entries of their own, and clears the fields that named it. Merging entries with Bulk Edit moves their senses and the references to them onto the surviving entry.
A FLEx or CLDF import keeps the sense structure of the lexicon, and so does every export of it. FLEx’s variants and complex forms come in when you tick Variants and complex forms on the review screen, as variantOf and components fields.
4.5. Multi-word expressions
Some lexical units are longer than a word: an idiom, a phrasal verb, a compound name. You can link such a multi-word expression to a single entry by gathering its words:
-
Shift+click each word. Or, from one of a word’s cells, press Shift+→ (or Shift+←) to take in its neighbours, holding Ctrl/Cmd as well to skip a word that does not belong.
-
A blue line shows the words gathered so far. Press Enter to link them, or Esc to give up.
-
The popover works as it does for one word: pick an entry, or create one. A new entry gets the words joined by spaces, which you can edit to the citation form, and the type multi-word expression (a phrase, to FieldWorks).
A linked multi-word expression draws as a line under its words, labelled with the entry’s form. A skipped word shows as a dotted stretch. A word can belong to several multi-word expressions at once, and each word keeps its own entry link and morphemes regardless.
Click a line’s label to change the entry, unlink it, or accept one that Auto-analyze made. While that popover is open, Shift+click a word to add it to the expression or take it out.
5. Auto-analyze
Most of the words in a running text are repeats, and much of the rest can be guessed at. Auto-analyze, a button on the Analyze tab’s toolbar, makes that first pass over a whole document so that your time goes into checking rather than typing. It runs up to four steps in order, each of which you can switch on or off. Everything it produces shows in violet until you accept or discard it (see Provenance), and nothing a person has written is ever changed.
5.1. The four steps
-
Propose translations. A translation service drafts a free translation for every sentence. It runs first because the later steps read the translation. Needs a connected translation service.
-
Copy previous analyses. Every word that has not been analyzed at all is given the analysis the same word form has most often had elsewhere in the project. This is built in and needs no service (see Copying previous analyses).
-
Propose segmentation and glosses. An analysis service, such as PolyGloss or a language model, proposes morphemes and glosses for every word a person has not analyzed. Needs a connected analysis service.
-
Link to the lexicon. Words and morphemes no person has linked are linked to lexicon entries, by the built-in rule described below or by a service. It runs last so that it can link the stems step 3 produced. On a word step 2 copied, the built-in rule keeps the links the copy brought and links only what has none. Needs the project to have a vocabulary.
A service step is unavailable when no service for it is online. An analysis service writes its glosses into the morpheme field whose lexicon field is the entry’s gloss, and a translation service into the sentence field whose role is the translation, or the field chosen in the dialog. With no such field the dialog asks for one.
A run over a long document takes minutes. The dialog closes when the run starts, and a banner at the top of the document lists the steps, ticks them off as they finish and shows the progress and elapsed time of the step under way. Editing is paused while the run writes. Stop on the banner ends the run at the step it is on, and what that step wrote stays. Leaving the document ends the run after the step under way, and reopening it says which steps did not run. Closing or reloading the tab asks first. When the run finishes, a message says what was done.
Then review the result with Ctrl/Cmd+Enter, which accepts a whole word, Ctrl/Cmd+Backspace, which discards its proposal, and Ctrl/Cmd+Shift+↑/↓, which jumps between the words that still have unconfirmed material.
5.2. Copying previous analyses
Step 2 reuses your own work, in the manner of FieldWorks' guesses from previous analyses. A word receives a copy only when it is unanalyzed: still a single morpheme identical to the word, with no annotation values and no lexicon link. A word you have segmented, glossed, or linked, even partly, is never touched.
For each such word, Plaid IGT looks at every other occurrence of the same form in the document and in up to 25 other documents of the project, those with the most words it is looking for, and copies the analysis given most often: the morphemes, the lexicon links, and the annotation values. If two analyses are tied, the word is left alone for you to decide. Only analyses a person made or confirmed count, so one machine pass never feeds the next. What a copy carries is set per project under Settings → Services (see Services).
5.3. Linking to the lexicon
The built-in linking rule of step 4 works word by word and morpheme by morpheme. A form that has been linked before follows precedent: the entry that form has most often been linked to in the project, counting only the links a person made or confirmed. A form never linked before is linked to the lexicon entry with the same form, if there is one. A whole word is never linked to an affix or clitic entry. The rule replaces only links nobody has confirmed: its own, links copied with an earlier analysis, and links to an entry that no longer has the word’s or morpheme’s form. A link a person made or confirmed stays, and so does an entry an analysis service chose for that occurrence. Runs of words whose joined form matches a multi-word expression entry are linked as one (see Multi-word expressions).
Alt+↓ on a morpheme field with a lexicon field, such as Gloss or POS, opens a list of each sense with its value, the alternatives an analysis service weighed with their probability, and the values the project has given the form. A click on the morpheme’s link opens the same list for one field: the field whose values an analysis service gave with their entries, else the field whose lexicon field is the entry’s gloss, else the first morpheme field with a lexicon field. Choosing a row sets the link and the gloss together. A row with no entry leaves the link as it is, and a sense with no gloss leaves the gloss as it is. Chosen in the popover, a sense keeps a value a person wrote that is not the linked sense’s own, and fills the other fields with a lexicon field that are empty or hold an unconfirmed proposal. A gloss typed by hand changes only the gloss.
|
Tip
|
Accept links as you go. An entry a form has been linked to becomes the precedent for every later occurrence of that form. |
6. Activity
A project’s Activity tab, for maintainers, shows who has been working and on what.
The upper table counts each person’s changes over a window you choose. A change is one action, so importing a corpus counts once, not once per sentence. Each row also counts the documents the person touched, with a bar for each day of the chosen window (the last 30 days for All time). Below it, the members who have made no changes in that window are named.
The lower table is the project’s history as a feed, newest first, each row naming the person, what they did, and the document it happened to. Changes made in this project’s vocabularies, such as merging, renaming or deleting entries, appear here. Changes made in a vocabulary linked from another project appear only in that vocabulary’s History and in that of each document they changed. For one document’s history, and for putting a document back the way it was, see History.
7. Settings
A maintainer opens a project’s Settings tab. It has five sections, listed down the left side.
7.1. General
The project’s name, its languages, the Research and Project plaid switches, and the option to delete the project. Deleting a project deletes its vocabularies with it, and is refused while another project links one of them.
Two languages are recorded: the object language, which the baseline text is in, and the meta language, which glosses and translations are written in.
Each has a name, an ISO 639-3 code, a Glottocode, and a writing system tag, the label FLEx and ELAN files carry on their text (oni, pmy, en).
An import from FLEx, ELAN, or CLDF fills these in from the file.
The languages go out with exports such as CLDF, where a dataset without a Glottocode cannot be linked to any other.
Under Research, Record responses to suggestions records when a suggestion is shown to a member, accepted or replaced, and when an assistant plan is expanded. It is off by default. Under Project plaid, Show the project’s plaid draws the project’s plaid beside its name and in the browser tab.
7.2. Text and vocab
Three things live here:
-
The extra orthographies each word can carry beside the baseline form, such as a phonetic transcription or a practical spelling.
-
Which vocabularies the project draws on (see Vocabularies). The project’s own are always linked, and New vocabulary makes another. A vocabulary of another project can be linked by a maintainer who is also a member of that project.
-
The project’s special characters: the backslash codes that type characters your keyboard lacks (see Typing Special Characters).
7.3. Annotation
The annotation fields at each scope, sentence, word, and morpheme.
You can add, rename, reorder, and remove them, and removing a field first tells you how many values it holds.
Each field records the language its values are in, as a writing system tag such as pmy or nl, which FLEx and ELAN exports label it with.
An import fills it in, and a field named with a tag in parentheses, such as Gloss (nl), is given that tag.
A word or morpheme field also records its lexicon field, the entry field that holds the same kind of value, such as Gloss for a gloss field.
Guesses from a linked entry and the list of senses under Alt+↓ read it, and a field with no lexicon field takes nothing from the lexicon.
A new field named like an entry field, such as Gloss or POS, is given that entry field, and an import fills it in.
A sentence field records its role: the translation, the literal translation, a note, or none.
Exports read it to know which field is the free translation, and services take the sentence field whose role is the translation as their translation field.
A new sentence field named Translation, Literal Translation or Note is given that role, and an import fills it in.
Tokenize new text, on by default, splits the text added on the Baseline and Media tabs, and by the assistant, into words.
The ignored tokens rule says which tokens carry no word-level annotation and host no morphemes, usually punctuation. Unicode punctuation takes every punctuation mark and symbol except ∅ and pictographs such as ★. Its exceptions are the characters that behave as letters, such as the apostrophe of an orthography that writes a glottal stop with one. Each listed character joins the word around it instead of splitting it, and stays on the word’s form in the lexicon. Explicit list instead takes only the tokens you list, separated by commas. Punctuation the rule splits words at, typed at the start or end of a word on the Baseline or Media tab, stays out of the word.
This section also holds the project’s tagsets (see Tagsets) and the document metadata fields shown on every document’s Details tab.
7.4. Access
The project’s members and their roles, invitation links, and API access. See Sharing and Permissions.
Review work, a checkbox per member, marks everything that person enters as contributed, in amber, until a verifier confirms it (see Provenance). It is independent of the role: tick it for a community member or a student whatever access they have, and leave it off for a trusted collaborator.
7.5. Services
Which service, or built-in method, each task uses by default: tokenization, transcription, speech detection, translation, analysis, and auto-linking (see Services). A default can carry default options too, and people can still switch per use. Each task also has default fields, the fields a service writes into or reads, such as the gloss field an analysis service fills and the translation it reads. Setup sets them from the fields it makes, and a renamed field stays the default. The Copy previous analyses card says whether Auto-analyze ticks its copy step by default, and what a copy takes from a word’s previous analyses: its segmentation, its links, its field values, or any combination.
8. Tagsets
A tagset is a controlled list of values for a field: the parts of speech your project uses, the Leipzig abbreviations you gloss with, the genres a document can be. Tagsets are defined under Settings → Annotation, and any field can then point at one. Several fields can share a tagset, so Gloss at the word scope and Gloss at the morpheme scope agree. A vocabulary’s entry fields take tagsets too, kept with the vocabulary (see Entry fields).
Each tagset has a mode:
-
Open: the whole list is offered while you annotate, but anything can be typed.
-
Closed: only values on the list can be entered. For a fixed inventory such as part of speech.
-
Closed, plus lexical glosses: a value is allowed if it has a lowercase letter or is in a script without capitals, and anything else must be on the list. This is made for glosses, where a stem’s gloss is a word like
dogand a grammatical gloss like1SG.NOMis built from tags.
In the third mode, a lowercase abbreviation beside a tag in the same gloss is read as a tag, so the sbj and pfv of sbj:3.pfv must be on the list.
The gloss of an affix, a clitic, or a zero morph is never read as a lexical gloss.
A gloss is composite, so a tagset lists the parts and names the delimiters that join them (for Leipzig, ., :, and >).
In a governed cell the list opens on its own and offers the part under the caret, and a value the tagset refuses is underlined in red.
A value can carry a description, shown beside it in the list.
Add values used in this project fills a new tagset with the values a field already uses, so an existing corpus comes under a tagset without retyping anything.
A tagset is enforced as you type and in Bulk Edit. A Closed tagset also refuses an outside value from an approved Assistant plan or a script. Values brought in by an import, or proposed by a service, are kept, but cannot be confirmed while they are outside the tagset. Closing a tagset, removing a value from a closed one, or assigning a closed tagset to a field is not saved while the field holds values outside the list that a person typed or an approved Assistant plan wrote. A Not saved notice names those values, with how many annotations use each, and for a composite value the parts that are not on the list. When a value is written in the moment before the save, a message names it instead. Change them first, or add them to the tagset. A change to one tagset is never held up by the values of a field that another tagset governs, nor by a field it is out of force on (see below).
The project’s Validation tab (maintainers) lists, for each governed field, every value in the project that its tagset refuses, with a count. It offers the remedies: Add to tagset, Fix in Bulk Edit, or open each occurrence and correct it by hand. A field under an Open tagset is listed too, with the values that closing it would refuse. It also finds zero morphs written some other way than ∅ (see Typing Special Characters).
A closed tagset is out of force on a field when values a person typed were already outside its list as the tagset reached the field. Validation then names the tagset and the field above the list. The tagset can still be changed, and each save names the values still in the way. Once those values are changed, Put in force applies the list. Saving any tagset or field does the same for every field whose values allow it. A save of tagsets or fields, in the project or a lexicon, made while another maintainer changes the tagsets is not saved, and says Tagsets changed elsewhere. Reload and redo your change. So is an approved Assistant plan of tagset changes, and one that assigns a tagset deleted or renamed since it was made.
Validation also names a field that allows more than one annotation on a word, morpheme or sentence, such as a field a script made, and a field that holds a list no closed tagset names, such as a field whose tagset was deleted. Under the field it lists each word, morpheme or sentence that holds more than one, with the document and the values, and a link to it in Analyze. Ignored tokens, which have no fields in Analyze, are not listed or counted. Analyze shows one of the values at a time, and clearing it shows the next. Put in force applies one annotation on each once no word, morpheme or sentence holds two, and Remove the list removes the list. A list whose tagset was made closed again in the meantime stays, and the notice says so. Neither changes an annotation.
|
Note
|
The Assistant and the LLM glossing service read a governed field’s tagset before proposing values, so they use its spellings. A service’s proposals are not refused, even outside a closed tagset. The Validation tab lists them. |
9. Guidelines
The Guidelines tab is the project’s own annotation manual: the conventions the people working on it have agreed to, written down where everyone can read them. Tagsets say which values a field takes. Guidelines are for what a list of values cannot say: that loanwords are not segmented, that a free translation is idiomatic rather than word-by-word, that this project follows Leipzig except for the ergative. Everyone who can open the project can read them, and anyone who can annotate can write them.
Each guideline has a title and a body written in a rich-text editor (bold, headings, lists, tables, links), and Markdown switches to the source. Name the subject rather than the rule, so that the title still fits after the rule is refined. If a title cannot name what a guideline covers, it is two guidelines. Open the body with the rule itself, since that first line is what the Assistant sees when the manual is too large to send whole.
Pinning a guideline sends it to the Assistant in full with every question. Unpinned guidelines are sent in full while the manual is small, and the Assistant opens them by title once it is large.
The Assistant can draft a guideline as well as read one. If you tell it a convention while asking about something else, it will propose writing it down, and you approve or discard it like any other change it proposes. When it changes a guideline that already exists, the card shows the passage it is replacing and what with. Every change to a guideline is recorded in the project’s Activity, with who made it and when.
Guidelines travel in the Plaid IGT archive and come back on import, so a project handed to somebody else arrives with the conventions it was annotated under.
10. Bulk Edit
The Bulk Edit tab (maintainers) changes many places at once, the way FieldWorks' Change Spelling does. You describe the change, preview every place it would touch, tick the ones you mean, and apply them as one action in the history. There are four operations:
-
Respell words: change the spelling of a word everywhere it occurs, keeping its morphemes, glosses, links, and other annotations. Optionally respell the same string inside morpheme forms and lexicon entry forms too.
-
Replace in a field: find and replace across one field’s values, as plain text or a regular expression, or across morpheme forms.
-
Re-analyze a word: give every occurrence of a word the same analysis, chosen from the analyses it already has in the project.
-
Merge lexicon entries: fold duplicate entries into one, moving every link and reference to the survivor.
Respelling a lexicon entry and merging entries need a maintainer of the project the entry’s vocabulary belongs to. Entries of any other vocabulary are listed and left as they are, as maintainers only, or as read only when its project is one you cannot open.
Respelling and replacing match the way Search does: a match contains the text, is exactly it, or matches a regular expression.
Contains ignores case, and keeps each match’s capitals when both boxes are typed in small letters: ka to ga makes Kalamang Galamang.
A change that would leave a word, a morpheme form, or an entry empty is skipped, and so is one a closed tagset refuses.
A long preview is paged, and a document’s heading ticks every match in that document. A place that changed after the preview is skipped, and the message after Apply says how many. When Apply stops partway, the message says what was written, and Apply again writes the rest.
11. Assistant
The Assistant tab is a chat with a language model that knows your project. Ask it what is still unglossed, whether a suffix is glossed consistently, or to summarize the morphology it can see. It reads the documents, the lexicon, the search results, the project’s guidelines, and the history to answer, and it shows what it read.
A line above each answer sums up the work behind it, such as "read 87 sentences in 40 documents · 2 searches · 42 steps". Opened, it lists each step the assistant took, and a step opens to show exactly what it read.
When you ask it to change something, it does not touch the data. It proposes a plan, which you approve with Approve as unverified or Approve as verified, or Discard. The plan lists every change under the document or lexicon it lands in, and each word links to its place in the editor, so a change can be checked before it is approved. A change that rewrites your baseline text is marked Rewrite, and one that changes something a person made or accepted is marked Accepted. Deleting a sentence is a rewrite: its text goes, with its words, their analyses and links, and its translation. A change over the whole project, such as replacing a gloss everywhere, is one row with its counts, and Show where lists the documents.
Approve as unverified records every value the assistant writes as its proposal, unconfirmed, as a service’s proposals are, to be confirmed later in the editor. Approve as verified records them as the assistant’s work accepted by you. A confirmation in the plan is made under either button. Tick Record as human-made to record them as your own work instead, and the card then offers a single Approve. A plan of only confirmations, deletions and other changes that write no value offers a single Approve as well. If your work is reviewed in this project, the second button reads Approve as your contribution, and what you approve with it waits for a reviewer, as anything you enter does. Once a plan is applied, its card says how it was approved, such as Approved as unverified. Applying a plan needs Writer access. Adding a lexicon’s entries and changing their fields needs a writer of the lexicon’s project, and renaming, merging or deleting them a maintainer of it or an administrator. The assistant does not change a lexicon linked from another project, whatever your role in the lexicon’s own project. Ask it from that project instead.
A plan is refused when a sentence it changes has been edited since it was made, and the card then reads Out of date. Ask again to have it planned on the current text. A plan that stops partway reads Partly applied, and the assistant can finish it. A plan that only keeps values as they are reads Nothing to write, and can only be discarded.
Where the assistant’s model runs, and where what you send it goes, depends on the Plaid site you use. When the people running the site’s assistant have said, the panel shows their statement at the start of a new conversation and behind the About this assistant button in its header. Otherwise, ask the site’s administrators.
The assistant picker names who runs each assistant. An assistant run by a maintainer of the project or by an administrator answers every member. One run by anyone else answers only the person who runs it.
A reply lands whether or not the page stays open, so you can leave, reload, or close the browser and find the answer when you come back. A conversation is worked on in one tab at a time. Opened in a second tab it is read-only, and Continue here moves it to that tab. Stop ends a turn early, even while the model is still working on its answer, and Retry sends the question again.
Conversations are private to you, though your server’s administrators can read them. They are kept per project, listed on the left of the tab, and each has its own link. All projects widens the list. A conversation can be renamed with the pencil beside it, downloaded or copied as Markdown, downloaded as a web page, or deleted.
11.1. Other projects
Where the assistant allows it, Add project, beside the paperclip, lets a question read other IGT projects you can open, to compare or to look something up there. Each project added is shown above the message box. The assistant only reads them, and plans changes in the conversation’s own project alone. A citation into another project carries that project’s name.
11.2. Files and PDFs
The paperclip in the message box attaches up to five files to a question, and dropping or pasting a file does the same.
The assistant reads tables (.csv, .tsv), text (.txt, .md), JSON, XML, .flextext, .eaf and .lift files, and PDFs.
Reading a .flextext, .eaf or .lift file does not bring it into the project. Import it for that.
A reply can also give you a file the assistant made, such as a cleaned table, to download.
To add many lexicon entries from a table, load it with Bulk Add as it is, then ask the assistant to tidy the new entries: merge repeated forms, normalize values, mark where they came from.
A PDF must have text in it, so a scan is refused. Ask about "§9" or "page 41" and the assistant reads that part. Text taken from a PDF can come out wrong where the PDF’s fonts are old, so check a form against the PDF before relying on it. Where the assistant can look things up on the web, it also reads a PDF from a link or a DOI pasted into a question.
11.3. The panel
The Assistant button in the header opens the same assistant in a panel beside whatever you are looking at, as does the handle on the right edge of the window. The panel stays open as you move around, and it can be dragged wider or narrower. It picks up the last conversation you had in the project you are in and keeps it as you move from document to document. The panel’s history lists your other conversations, and Open in Assistant moves the open one to the Assistant tab. An answer that ends while you are in another tab or window is announced with a short sound, a ● at the start of the tab’s title, and a notification if you allowed them when you first sent a message.
A question asked on a document is about that document, unless it names another. Ask, beside Copy on a sentence, puts that sentence into the question. Typing @ in the message box names a sentence of the open document, an entry of the vocabulary you are on, or a document of the project. Sentences are matched on what they say, so @ followed by a word finds the sentence that contains it.
11.4. Conversation size
The panel’s header shows how full the conversation is, and hovering over the bar gives the details. A conversation fills up as it grows, because every turn sends the whole thread again. Past 85 percent the bar turns amber. A full conversation sends nothing more, and New conversation beside the message box starts one with what you had typed.
11.5. Assistant availability
The assistant is a service that your server’s operator runs and picks the model for. In a project with none connected, the tab, the button, the handle and Ask are not shown. On the project list the button is shown whether or not an assistant is connected, and the panel says so when the project you pick has none. Conversations already saved still open from their own links.
12. Assistant jobs
An assistant job has the assistant work through documents on its own, by written instructions, with nobody at the screen. It writes the changes it decides on, and lists changes to the project’s setup for a person to make. A job can take hours, and it goes on after you close the browser.
Set up job, beside New document on the Documents tab, on the Jobs tab and in a document’s header, needs Writer access. It opens the assistant on a new conversation for setting the job up (see Setting up a job), on the Assistant tab when the window is too narrow for the panel. In a project with no assistant connected it opens the Start job form instead.
The form is also on the Jobs tab’s ⋯ menu, as Start without the assistant. It asks for the documents (all of them, or a selection, with the open document chosen when started from one) and the Instructions, which the job needs. Translate, Segment and gloss and Segment, gloss and link fill the instructions with a starting text to edit. A job takes at most 500 documents, so a larger project starts one on a selection: Select shown selects every document the search above the list shows.
A project has one job at a time. Jobs run one after another across the server, and a waiting job’s page says how many are ahead of it.
12.1. Setting up a job
In the conversation Set up job opens, say what the job should do, such as glossing a set of texts overnight. The assistant reads the documents, says what is empty in them, and asks about what it finds there, such as a word people analyzed in two ways. It can also analyze the first few sentences as an ordinary plan for you to correct first.
It then drafts the job as a card in the conversation: the documents, what the job Fills, whether its Text may be edited, and the Instructions, which you can edit. To change what the job fills or whether it may edit the text, tell the assistant, and it drafts a new card. Start job on the card starts the job, and Discard drops it. The card keeps how it ended, with a link to the job’s page once the job has started.
The job is planned to fill only what its card names, and edits the text only when the card says it may. Its page says so from the start, with a link to the conversation for the person who started it. A job started from the form has no such limits.
12.2. What a job writes
A job first plans how to cut the work: a few sentences at a time, one document at a time, or all the documents at once. Its plan also names what the job fills: segmentation into morphemes, any of the project’s fields at word, morpheme or sentence level, any of its orthographies, or nothing, for a job that only checks or counts. Its page says which, such as Fills: segmentation, Morph Gloss or Fills nothing. A word, a morpheme or a sentence is empty for a field when the field holds no value on it, and a word is empty for segmentation when it has no morpheme analysis. Worked a few sentences at a time, it takes every sentence, unless its plan takes only the sentences with something empty of what it fills. When planning ends with no plan, the job ends as failed without writing anything, and its page says Planning ended with no plan. When the assistant’s model fails during planning, the job is planned once more, and when it fails again the job ends the same way and its page says The assistant’s model failed during planning. Each piece is worked by the assistant with the project’s guidelines and the instructions. A piece of a few sentences is told what is empty in it of what the job fills, how people in the project and the job itself filled the same forms, and, when what it fills is linked to a lexicon, the lexicon entries spelled inside them. Its tools are those for setting values and reading the lexicon and earlier analyses, those for analyzing words, linking them and creating entries when the job fills those, and those for retyping, splitting, merging and deleting sentences when its plan allows text edits. A piece of a whole document, or of all the documents, has the assistant’s usual tools and how people in the project analyzed the same words.
A job writes every change it plans in its documents and in the project’s lexicons, in the name of the person who started the job: values, analyses, links, comments, changes to entries, and deletions. It can change and delete what people wrote or confirmed. It edits the text, retyping, splitting, merging or deleting words and sentences, only when its plan says the instructions ask for that, and its page then reads Text may be edited. Each value it writes is a machine proposal (see Provenance), also where it replaces a value a person wrote or confirmed. An analysis that keeps a word’s morphemes, as many and with the same forms, writes only the values that differ: every other value and link stays as it was, with who wrote it. An analysis of another shape replaces the word’s analysis. A job never deletes a document, and never changes the project’s setup: a guideline, a tagset, a document’s name or metadata, or a new document is listed in its report as a proposed change, for a person to make. So is a change in a document other than the one being worked (when the job works all its documents at once, in a document outside the job), a corpus-wide change, a confirmation, a restore of a document, and the renaming, merging or deleting of an entry when the person who started the job is not a maintainer of the project. When someone changes a sentence while the job works on it, the job does not write over the change: it takes the sentence again, and lists it as Changed during the job when it changes once more.
When it finishes, it makes its own analyses consistent: a word it analyzed in more than one way takes the analysis people gave that word, or else the one it gave at least two times in three.
The job keeps out of the way of people. It waits while anyone’s assistant or service run is using the AI model. A document someone is editing is left for later and taken up again after the others, and editing stays open while a job works in it.
12.3. Following and stopping a job
The project’s Jobs tab lists its jobs, and each opens to its own page. The page shows what the job is doing, how it cut the work and what the assistant told each piece of it (with a line when that was cut for length), each document’s progress, and once the job has ended its report. In the table of documents, a document the job did not finish reads Stopped partway once a sentence of it was finished, and once the job has ended each document has how many places it left blank, how many of its changes replaced or removed what people wrote, and how many changed the text, beside the links to what the job changed in its History. The report opens with the totals: what the job wrote and deleted, the values it set in each field, how many changes replaced or removed what people wrote and how many changed the text, and how much of each thing it fills is still blank, in the sentences it worked, in those it did not reach and in those it skipped. Then come the assistant’s summary of the job, the documents its plan left out, the changes it proposed, a note from each piece of its work, any tables it counted, the sentences it skipped and why, and what it left blank, each word or sentence with what it is blank for (or that a document could not be read for it), with what a refusal kept it from doing and how much is blank in the sentences it did not reach and in those it skipped. The report ends with the analyses it made consistent, the entries it created, and those of them with no links. A job that wrote no change, proposed none and made no table says so in one line, with no summary. A document the job is working in shows a banner with its progress.
Stop, on the job’s page or on the banner, is for the person who started the job and for maintainers. The job stops after the sentences in hand, and what it wrote stays.
12.4. Discussing a job with the assistant
Discuss with the assistant, on the page of a job that has ended, opens the assistant panel on a new conversation about that job. The assistant reads the job’s report and tells you what the job did. A job’s record holds what it did, the model’s note on each piece of its work, and nothing more of why. The report is as the job ended, so the assistant reads the documents as they are now before it plans a change, and you approve the change as any other plan. This is how a job’s proposed changes are made: ask the assistant for the ones you want. From any page, the assistant can also list the project’s jobs and read their reports.
12.5. Machine draft entries
An entry a job creates has the Status Machine draft, and a person’s edit of it leaves that status as it is. On a vocabulary’s Entries tab, N machine drafts lists only those entries, with any search on top, and its address can be shared. While it is on, Mark reviewed sets the status of every listed entry to reviewed. Delete, for a maintainer of the vocabulary’s project, deletes them all with their links, and leaves the glosses. A job’s report links to each lexicon’s machine drafts. An existing entry a job edits keeps its status. An entry a job created that ends with no links is listed in its report under Entries with no links, for a person to delete, and a job deletes such entries only when its instructions say so.
12.6. Undoing a job
Each document’s History has one entry for the job’s work in it, named Assistant job: with the first words of its instructions, and one for making its analyses consistent. A job’s page links each document to them, and each opens on what the job changed there (see Changes in an entry). Once a job has ended, Undo job on its page, for a project’s maintainers, takes back what the job wrote in every document and in the lexicon, as one entry in the history, and leaves what people changed since. A dialog lists what will be undone and what is left before anything is written, and a job too large to undo at once is undone document by document from each document’s History. Restoring a document to the moment before the job (see History) also takes back the job’s work in that document, but not in the lexicon.
12.7. The job worker
Jobs run only while the server’s operator runs the job worker, plaid-job-worker, with an AI model.
When no worker is running, a job waits, and the Start job form and the job card say so.
13. Services
Plaid IGT can call external services for certain tasks. A service is a small program, often a Python script wrapping a model, that connects to your project when it starts and advertises which task it handles. The AI assistance described in the introduction is connected this way. Once connected, it appears wherever its task is offered, and a maintainer can make it the default for that task under Settings → Services. Some tasks ship with a built-in method that needs no service at all.
Every service run works the same way. A button in the header of the panel its results land in opens a dialog with the Method to use, that method’s options, and the button that starts the run. The dialog closes when the run starts.
A run that writes to the document makes it read-only until it finishes, and only one such run goes at a time. A banner at the top of the document shows the run on every tab: its progress, what it is doing and how long it has been going. Its button spins until the run ends. A run also outlives the page that started it: reopen the document and the run is picked up where it got to. Stop on the banner ends a run, and whatever it wrote stays. Speech detection writes nothing, so it can run while you work, and its dialog shows its progress and Stop instead of the banner. A run that asks a model about each sentence stops early when the model gives no answer for two sentences in a row.
There are six integration points:
-
Tokenization: splits the baseline into words, on the Tokenize tab. The built-in splitter is always available. A service can split sentences as well (the bundled example uses Punkt).
-
Transcription (ASR): transcribes and time-aligns a recording, on the Media tab. There is no built-in, so this needs a connected service (the bundled example uses Whisper).
-
Speech detection: proposes where the utterances are in a recording, on the Media tab. The built-in runs in your browser (see Speech detection).
-
Translation: proposes a free translation for each sentence, as step 1 of Auto-analyze (see Auto-analyze). The bundled example uses a language model.
-
Analysis: proposes a morpheme segmentation and glosses for every word, as step 3 of Auto-analyze. There is no built-in. Three examples are bundled: PolyGloss, a multilingual glossing model, CWoMP, which learns a lexicon of morphemes from the project’s own analyses and gives its confidence for every morpheme, and a language model given the project’s own analyses as examples. CWoMP is trained on the machine that runs it, and morphemes the project confirms after training are used without training it again.
-
Auto-link vocabulary: proposes vocabulary links for unlinked words and morphemes, as step 4 of Auto-analyze. The built-in rule follows your project’s own precedent.
The Assistant is a service too (see Assistant). Whatever a service produces is marked machine-made, in violet italic, until a person confirms it.
Writing a service of your own is not hard. The Services chapter of the Plaid manual explains how.
14. Import and Export
14.1. Importing
Start a project from existing data on the New project screen. Five formats are supported:
-
FieldWorks (FLEx) backup (
.fwbackup): a whole FLEx project, with its texts, words, morphemes, glosses, translations, the full lexicon, and what each text’s Info tab holds. The backup’s writing systems become the project’s languages and each field’s language, so a later FLEx export is tagged the way FLEx expects. Words that cannot be aligned to their baseline are listed before the import and skipped. -
FieldWorks (FLEx) interlinear texts (
.flextext): one or more files exported from FLEx with File > Export Interlinear, with their texts and analyses but no lexicon. A.flextextholds only what the Interlinear view showed, so turn on every line you want before exporting. An analysis FLEx marks as a guess arrives as machine work to review. -
CLDF (
.zip): a Cross-Linguistic Data Formats dataset. Its examples become interlinear texts, its lexicon a vocabulary, and its language the project’s identity. -
ELAN (
.eaf): a set of annotation files in which each tier means the same thing in every file. Each file becomes a document, with its tiers, speakers, time alignment, and its recording when you choose that alongside. -
Plaid IGT archive (
.zip): recreates a project exactly, with its vocabularies, media, comments, and guidelines. Which members' work is reviewed is not carried, since it names people.
A FLEx backup keeps the times of sentences that were timed against a recording, and they become the segments on each document’s Media tab. To bring the recordings too, choose them together with the backup, and each goes to the texts that name it. A text whose recording is missing or too large is imported with its times. Upload the recording on its Media tab afterwards.
A FLEx part of speech has a name in each analysis language, v in English and kt.kerja in Indonesian, and only one of them is imported.
Choose which under Part-of-speech language on the review screen.
The review screen of an ELAN import has three parts, in the order you work.
Files lists each .eaf with the document it becomes and its recording under it.
Tiers is where each tier is given its role, and a tier is matched across files by its name, so files need not have the same tiers in the same places.
Preview draws a sentence as the interlinear lines the import would write, and redraws it as you change a role.
A file that names several audio recordings, one per speaker, has those chosen mixed into one recording with a channel each, as under Per speaker on the Media tab, at the quality chosen under Files. When one of those chosen is a video, nothing is mixed: the first one chosen, in the file’s order, is the recording, and the others are not used. A time origin set on a recording in ELAN is applied, so the segments stay on the speech they were aligned to.
The roles a tier can take:
-
Sentences: the transcription. Each annotation becomes one sentence, and their text, joined in time order, becomes the document’s text. Every import needs one such tier.
-
Words and Morphemes: tiers below it, dividing each sentence and then each word. A Words tier’s words are kept as they are in the file, punctuation included. Without one, each sentence is split into words as Tokenize splits it, with punctuation left out of the words.
-
Morpheme type: a tier naming each morpheme’s type in FLEx’s terms (
stem,suffix,enclitic). -
Time alignment: the segments on the Media tab. Leave it out and the sentence tier’s own times are used.
-
Orthography: an alternate spelling of each word, beside the baseline.
-
Sentence field, Word field, Morpheme field: one value per sentence, word, or morpheme, in an annotation field.
Tier names such as Translation-gls-nl, as a corpus prepared for FieldWorks names them, are read for their kind and language, so such a tier is offered the matching field.
A file made by ELAN’s Import FLEx is read the same way, with its fields named as a FLEx import names them.
When several tiers hold the sentences, a tier that names no participant takes its speaker from its name: W-Spch and K-Spch are speakers W and K.
ELAN files can also be imported into a project you already have, with New document → Import from ELAN on its Documents tab. Each tier is written into a field the project already has, or into a new field of its own. A file imported before says so on its row, and you choose whether to keep that document, replace it, or add a copy beside it.
An import that stops before it finishes, whether you stop it, it fails, or the tab is closed, leaves a project that does not open. Wherever it would open, it reads This import did not finish. Delete it and import again., with Delete for a project maintainer. A vocabulary the import made does not open either, and cannot be linked into another project, until the import is deleted. Delete asks for the project’s name, and deletes the project with the vocabularies the import made in it. It is refused while another project links one of them. A lexicon of another project you chose to import into stays, with the entries the import added to it. Nothing deletes it on its own, since the import may still be running in another tab. An import that finished but could not be marked done says so on the import screen, with Try again.
14.2. Exporting
Exports are driven by presets. A preset fixes a format and which orthographies, fields, and options it includes, so the same export can be run again later, or by someone else. Maintainers create presets on the project’s Export tab. Anyone can run one from there or from a document’s Export tab, on this document, selected documents, or the whole project. Only a Plaid IGT JSON archive carries comments.
Six formats are supported:
-
Plain text: an aligned, human-readable interlinear rendering. Its Word line setting prints the words as written, segmented into morphemes, or both, which is the four-line hand-in a field methods course asks for. Number sentences numbers them, and Document header prints the document’s name and metadata above its text.
-
FLEx (
.flextext+.lift): the texts as FieldWorks' interlinear XML (.flextext) together with the project’s vocabularies as a LIFT lexicon (.lift), zipped together. In FieldWorks, import the lexicon first, from the Lexicon area (File > Import > LIFT Lexicon), keeping the.lift-rangesfile beside the.lift. Then import the texts from the Texts & Words area (File > Import > FLExText Interlinear). The order matters because the texts link to entries that must already be there, and a README in the.ziprepeats these steps. -
CLDF TextCorpus (
.zip): a Cross-Linguistic Data Formats dataset, one CSV per table with acldf-metadata.jsondescribing them. CLDF has no place for vocabulary links, the marks that tell machine work from a person’s, or most morpheme types, so these are left out, and the preset lists them before the export runs. -
ELAN (
.eaf): one annotation file per document, bundled into a.zipwith its recording when it has one. Segment words into morphemes decides whether the tiers reach below the word, and One tier set per speaker gives each speaker their own tiers. A segment that crosses a sentence break is left out, and the export says how many. -
LaTeX book (
.zip): LaTeX source for a book of the texts, to compile with LuaLaTeX. See LaTeX book. -
Plaid IGT JSON (
.zip): a lossless archive you can import again later.
An export holds everything its preset names, or it fails, and the export panel says what could not be read.
A FLEx preset also carries the writing-system tags FieldWorks will see, for the baseline text, each alternate orthography, and glosses and translations. Use the codes your FieldWorks project already uses, or FLEx will offer to add a new writing system when you import. Each field goes out under its own language where the project knows it, and the box beside each field overrides it. A new FLEx or CLDF preset sends the sentence field whose role is the translation as the free translation, and a CLDF preset sends the one whose role is a note as the comment and the morpheme field whose lexicon field is the entry’s gloss as the gloss.
14.2.1. LaTeX book
A LaTeX book export turns the documents in its scope into a book of interlinear texts: a title, a table of contents, a list of abbreviations, and one chapter per document.
Each sentence is a numbered example with the lines the Analyze tab shows, then the free translation and the other sentence fields.
Each text begins on a new page under its name, and a punctuation mark is set with the word it belongs to.
The lines of an example print every character as it was typed, -- as two hyphens included, while the translation and the other sentence fields are set as prose, with curly quotation marks and -- as a dash.
The preset’s Example lines and Sentence fields lists turn each line on or off and set their order.
Document metadata lists each document’s metadata under its chapter heading.
Glosses written in capitals, such as PL or 1SG, are set in small caps.
The list of abbreviations names every one the texts use, with the description its tagset gives it or the meaning from the Leipzig Glossing Rules, and a blank to fill in where there is neither.
A document written right to left is set right to left.
With Vocabulary chapter on, the book ends with the project’s vocabulary, either the entries the texts link to or every entry. Each vocabulary, and each of its fields, can be switched off. Each headword shows its form, its number, its gloss and fields, the examples promoted to it, and then each sense. In the PDF, each linked word in the examples links to its entry.
The .zip holds:
-
main.tex: the book, with a marked place for front matter such as a title page, preface, and acknowledgments. -
abbreviations.tex: the list of abbreviations. -
texts/: one file per document. -
vocabulary.tex: the vocabulary, when the book has one. -
latexmkrcandREADME.txt: the compiler setting and how to compile.
To make the book, unpack the .zip and compile main.tex with LuaLaTeX through latexmk:
latexmk -lualatex main.tex
Run by hand, LuaLaTeX has to run twice, or the table of contents stays empty.
An online LaTeX editor that compiles with latexmk takes the .zip as it is, with its compiler set to LuaLaTeX.
The book uses the Charis SIL font, with a Noto font for any other script the texts are written in.
The look of each line, such as italic words, is set by the \Plaid... commands at the top of main.tex.
To grab a single sentence instead, use Copy as IGT in the Analyze tab.
15. Search
A project’s Search tab looks across every document.
Pick where to look: word forms, morpheme forms, any annotation field at any scope, or the lexicon’s entries.
Type what to find, and choose whether a match contains it, is exactly it, or matches a regular expression.
Contains ignores case, and every kind finds text however its accents were typed.
An accent searched for on its own, such as a tone mark, also finds it inside a letter typed with it, so a lone acute finds the one in á.
In a regular expression the accent may also be written as an escape, such as \x{301} for the acute, and \p{M} finds every accent, inside letters too.
A regular expression is read the same way in Search, in Bulk Edit, in a vocabulary’s Replace, and by the Assistant.
\w is a letter, mark, digit, or underscore in any script, \b the edge of a word, and \p{IsArabic} any character of the Arabic script, and likewise for every script.
(?i) at the start ignores case.
In a replacement, $1 is the first group and $& the whole match.
A replacement cannot hold a backslash, so an accent goes in as the character itself.
Results come in two views:
-
Hits lists every match grouped by document, each in its sentence with the translation. Click one to open that sentence in the Analyze tab.
-
Frequencies counts the distinct matching values instead. This is how you see every gloss a suffix has been given, or every spelling a word has had.
16. History
Every change in Plaid IGT is recorded in an audit log, with a meaningful message and the identity of the person who made it. Past states of any document can be viewed, read-only, with each vocabulary entry as it read at that moment. Comments are not part of the history, and neither is the recording itself, so a past state cannot bring back a deleted recording.
Open a document’s history with History at the end of its tab row, or from the rail at the left edge of the editor.
Each entry is one action as you performed it, such as "Merge morphemes" or "Tokenize".
An edit made in a cell names the field, the word or morpheme, the sentence and the value, such as Gloss of "dogs" in sentence 3: DOG.
Selecting an entry shows what it changed (see Changes in an entry), and Document above it shows the document as it was right after that action.
An entry made of several changes expands to show each one, and selecting one of those shows what that one change did, with the document right after it under Document.
Actions performed by a service show up the same way, under the service’s name and naming the person who ran it: LLM translation (12 sentences), requested by Ana.
A change made through an API token names the token too.
Some entries are marked with their kind:
-
Import: a file brought in.
-
Automatic: a run of a service or of a built-in step, such as Auto-analyze’s.
-
Assistant: an approved Assistant plan.
-
Bulk edit: a change made in Bulk Edit or a vocabulary’s Replace.
-
Guess taken: a faint guess accepted.
-
Review: a proposal accepted as it stands.
-
Repair: a repair of the document’s data, rather than an edit.
To bring a document back to an earlier state, select that entry and click Restore. Everything in the document goes back to how it was at that moment, except the recording. A dialog lists what will change before anything is written, and says what cannot come back, such as a field deleted since. The restore is one more entry in the history, and the message that confirms it offers Undo. Restoring is for a project’s maintainers.
To take back what one entry did, and nothing else, click Undo over its changes. Each value, word, morpheme and link it wrote in the document goes back to how it was just before the entry, as one more entry in the history, which can itself be undone. A dialog lists what will be undone before anything is written.
What cannot go back is left as it is, and the dialog lists it with the reason and where it is. A value changed again later is left, and so is anything something was added to since, such as a word that now has a gloss. Lexicon entries are not undone from a document: undoing the same entry in the vocabulary’s History puts back its entries and every link to them, such as the links a merge moved. Changes to settings, fields, vocabularies and members are never undone. Undoing is for a project’s maintainers.
16.1. Changes in an entry
An entry opens on the sentences it changed, in document order, each drawn interlinear with the value after the entry above the one it replaced, struck through. Each morpheme’s values are shown apart, so a gloss written beside another in the same word is marked on its own. A word segmented anew shows its old morphemes struck through on a line under the new ones. Only what the entry itself wrote is marked: a value another change wrote on a morpheme the entry made reads as it is. Each value carries its usual mark, so a machine-made, contributed or confirmed value reads as it does in the grid, and a person’s value the entry replaced is struck through in red. A person’s value counts as replaced when the entry deleted it, changed it, or wrote it again as a machine’s. A change to the text is shown as old and new text and is never a person’s value replaced. A word or sentence the entry added or deleted shows as such, and a sentence someone changed after the entry says Changed again later.
The count above the sentences says how many changes the entry made and how many replaced a person’s value, and A person’s value replaced shows only the sentences where one did. Below the sentences come the document’s name and metadata, the lexicon entries the entry created, changed, merged or deleted, each a link, and counts of changes the sentences do not show: to transcript segments, to other annotations (those of other apps, and changes the grid draws no difference for), and to annotations deleted later. Every change in an entry is read before it is drawn, with a count of those read so far. The sentences are drawn 50 at a time, and Show more sentences draws the next. An entry too large to draw reads Too many changes to show here., with Show document for the document as of it. The address of the page names the entry, so it can be shared, and a job’s page links each document to its entries.
16.2. Vocabulary history
A vocabulary has a history of its own, opened with History beside the vocabulary’s name. It lists every change to the vocabulary and its entries, but not links, which belong to each document’s history. Selecting a history entry shows what it changed: the entries it created, changed, merged or deleted, before and after, and the links it moved, removed or made, counted per document. Links are not counted for an entry with too many changes in documents to count. Vocabulary above it shows the vocabulary’s entries as they were at that point. History on an open entry lists only the changes to that entry.
To put one entry back as it was, open it in the past state and click Restore. An entry deleted since comes back whole, but without its links in documents. Restoring an entry is for maintainers of the vocabulary’s project.
Undo over a history entry’s changes takes back what it did to the vocabulary’s entries and to every link to them, in any document. An entry it merged or deleted comes back with its links, and an entry it created goes. As in a document, the dialog lists what is left and why, and undoing is for maintainers of the vocabulary’s project.
17. Sharing and Permissions
Projects are shared by adding members, each with a role:
-
Reader: can view everything but make no edits, comments included.
-
Writer: can edit and delete documents, link vocabulary, and add and edit entries of the project’s vocabularies.
-
Maintainer: can also change project settings, manage members, and delete the project and its vocabularies.
Add and manage members under the project’s Settings, in Access. Members are found by name or by email address. When you only have Reader access, the editor shows everything, but a banner says so and its controls are inert.
Invitation links let people join without an administrator creating each account. A maintainer mints a link that grants a role on the project, good for a set number of uses and a set number of days. Whoever opens it chooses their own password and lands in the project. Someone who already has an account signs in from the link instead and joins the project with the link’s role. A link is not addressed to anyone, so record who it was meant for in its private Label. A link can be revoked at any time, and it stops working if its maker loses the access it grants.
An administrator can also create accounts directly from the same Access section, each added to the project with the role chosen, and can mint a password-reset link for someone who is locked out.
Scripts and services should not borrow your sign-in. Mint a named API token on your profile page instead. It can be revoked on its own, and its name appears in the history. A token cannot manage accounts or passwords, create or revoke tokens, or make invitation and password-reset links.
18. Administration
An administrator sees an Admin link in the top bar, covering the whole server rather than one project.
-
Users is the account directory: create, edit, deactivate and reactivate accounts, and mint a password-reset link. Deactivating signs the person out and takes away their project roles and API tokens, but leaves their work and their name on it.
-
Invites is every invitation link on the server. It can also mint a numbered set of single-use links in one step, and a link with No project creates an account and grants nothing else.
-
Activity is the same view described under Activity, across every project.
-
Projects lists every project on the server, with a way in for an administrator who was never added.
-
Vocabularies shows which projects use each vocabulary, and which vocabularies no project uses.
-
Services shows which services are running across every project. A service that has gone for good can be forgotten here.
-
Assistant is every conversation anyone has had with the assistant, readable and downloadable.
-
Server reports the version, the database, the backups, and which documents are open for editing. From it you can take a backup now, release a document someone left open, or clear a sign-in block for someone who mistyped their password too many times.
-
Logs is what the server has been doing since it last started: the requests, with the account behind each one, and the warnings and errors. Live keeps the page up to date while a problem is happening.
19. Typing Special Characters
Most language data needs characters a keyboard does not have. In any field that holds language data, type a backslash followed by a two-character code, and the code turns into the character as you finish it:
| Code | Character | Name |
|---|---|---|
|
ə |
schwa |
|
ŋ |
eng |
|
ʔ |
glottal stop |
|
æ |
ash |
|
ː |
length mark |
|
ˈ |
primary stress |
|
∅ |
zero morph (see below) |
The codes are Praat’s, so if you already use Praat they are the ones you know.
There are over four hundred of them, and the full list is under Settings → Text and vocab → Special characters.
For anything not in the list, \u0250 and the like give any character by its Unicode number.
Type two backslashes for a plain backslash.
Codes work in the Analyze grid, the baseline, the transcript, lexicon entries, and document metadata. They do not apply where a backslash has to mean itself, such as a regular expression in Search or Bulk Edit.
Every code is the project’s to change. Under Special characters, search for a code to see what it types, point it at a different character, remove it, or add a two-character code of your own. A change applies to everyone working on the project, and the reset button beside a changed code puts it back.
19.1. The zero morph
A morpheme with a grammatical job and no sound is written with the empty set sign, ∅.
It has a code of its own (\00) and a shortcut in the Analyze grid: Alt+0 in a morpheme form types it.
Write it as ∅ rather than a capital O with a stroke or a zero, so it exports as a zero morph.
The Validation tab lists any morpheme that looks like a zero written another way.
|
Tip
|
If you type a lot of one language, a system keyboard such as SIL Keyman is worth setting up. It works everywhere on your computer rather than only here. |
20. Keyboard Shortcuts
Every shortcut that takes Ctrl also takes Cmd on a Mac, and Alt is the Option key there.
The tables below give the shortcuts as they come.
You can change most of them under Profile, in Keyboard shortcuts: choose Change beside an action and press the keys you want.
A change follows your account to any computer you sign in on, and shows in the ? legends.
Reset puts one shortcut back and Reset all puts back every one.
Enter and Tab, each with or without Shift, the arrow keys, Escape, Backspace, Delete, and - and = in a morpheme form cannot be reassigned.
20.1. Any screen
| Shortcut | Action |
|---|---|
/ |
Focus the search box on the screen, when it has one and you are not typing in a box. |
20.2. Analyze grid
| Shortcut | Action |
|---|---|
Enter |
Move to the next cell in the row, accepting the faint guess if the cell shows one. Of the keys that move one cell, only Enter accepts a guess. |
Tab / Shift+Tab |
Move to the next or previous cell in the row, leaving a guess as it is. Shift+Enter moves back the same way. |
↑ / ↓ |
Move between rows. ← / → move along the row from the ends of a value, following the words: in a right-to-left sentence ← moves forward. See Details. |
Esc |
Cancel the edit and leave the cell. With a value list open, close the list first. In a cell with nothing typed, Esc lets the next Tab leave the grid, and Shift+Tab go back before it. |
Alt+↓ |
List every value the project has given this form, with counts. On a morpheme field with a lexicon field, list the senses with their values. ↑ / ↓ move through the list, Enter takes the highlighted row, Esc closes the list. A field held to a tagset opens its list on its own. |
Enter (sentence field) |
Save the value and leave the field. Shift+Enter starts a new line, and Tab moves to the same field of the next sentence. |
- |
In a morpheme form, split the morpheme at the cursor. |
= |
In a morpheme form, split at the cursor as a clitic boundary. The clitic is typed by its place in the word (see Morpheme types). |
Backspace |
At the start of a morpheme form, merge it into the previous one. The merged-in morpheme’s gloss and other values, its lexicon link and its comments are not kept. In an emptied morpheme form other than a word’s first, delete the morpheme with its gloss and other values. |
Alt+- / Alt+= |
Type a literal hyphen or equals sign in a morpheme form. |
Alt+0 |
In a morpheme form, type the zero morph ∅. |
\ + a two-character code |
Type a character the keyboard lacks, anywhere language data is typed: |
Ctrl/Cmd+Enter |
Accept the current word’s whole analysis, every faint guess in its cells included, and jump to the next word. On a sentence field, accept its proposed value and jump to the next sentence. A value not in its closed tagset accepts nothing and keeps the caret on that cell. |
Ctrl/Cmd+Backspace |
Discard the current word’s unverified proposal (its machine-made morphemes, glosses, and links) and jump to the next. Anything a person made or confirmed stays. On a sentence field, discard its proposed value. Delete does the same as Backspace. |
Ctrl/Cmd+Shift+↑ / ↓ |
Jump to the previous or next word or sentence field that still has unverified proposals. |
Ctrl/Cmd+↑ / ↓ |
Move between vocabulary links awaiting your review. On one, Enter accepts it and Backspace removes it. |
F2 |
Respell the current word in a box over it, its text selected. Enter saves, Esc or clicking elsewhere cancels. The same action is Respell in the word’s popover. |
Shift+click a word |
Gather it into a multi-word expression (again to take it out). Shift+← / Shift+→ from a word’s cells do the same along the sentence, and Ctrl as well skips a word. |
Enter (while gathering) |
Open the entry popover to link the gathered words to one entry. Esc drops them. |
Enter / Space (row label) |
Open the menu that picks which rows show. Esc closes it. |
↑ / ↓ (entry popover) |
Move through the entries. Enter links the highlighted one, on a morpheme together with its gloss, accepts a proposed link, and on a word unlinks the linked entry, Shift+Enter links it to every unlinked occurrence of the form in the text, on the create row Ctrl/Cmd+Enter creates the entry as typed, Ctrl/Cmd+Shift+Enter does both, Esc closes the popover. |
Ctrl/Cmd+Enter (comment popover) |
Post the comment. Esc closes the popover. |
20.3. Tokenize tab
| Gesture | Action |
|---|---|
Drag |
Make a word of plain text, or merge the words dragged across. |
Click a word |
Split it at a character. Esc closes the split points. |
Right-click a word |
Delete it. |
Ctrl/Cmd+click a word |
Split its sentence there, undoing a merge. |
Alt+click a word |
Open it on the Analyze tab. |
The arrow button beside a sentence’s number merges it with the previous one.
20.4. Media tab
| Shortcut | Action |
|---|---|
Space |
Play or pause the recording, when you are not typing in a box. |
Shift+Space |
Play or pause the segment you are in, or the selected stretch of the timeline. In a time box too. |
Shift+← / Shift+→ |
Move playback one second back or forward, in a row or out of one. |
Shift+↑ / Shift+↓ (transcript) |
Play faster or slower, a quarter of normal speed at a time. |
Alt+↑ / Alt+↓ (transcript) |
Move to the row above or below, from anywhere in the row. |
Enter / Esc (transcript row) |
Save the row and move to the next, or put the row back the way it was. |
↑ / ↓ (transcript row) |
Move to the row above or below, from the start or end of the text. |
Enter (proposed row) |
Make a segment of the proposal, with the text you typed. Moving on saves it too. |
Tab (transcript row) |
Move to the next field in the row. |
Alt+Enter |
Split the segment at playback, its text between the words at the caret. Outside a row, the segment playback is in. |
Alt+M |
Merge the segment with the next one. |
↑ / ↓ (segment time) |
Step the digit box under the caret: milliseconds by 10, or 100 with Shift. ← / → move between boxes, Backspace zeroes the box. |
Ctrl/Cmd+Enter / Esc (new-segment popover) |
Save the segment, or cancel it. |
Esc (timeline) |
Clear the selected stretch. |
Ctrl/Cmd+wheel / wheel (timeline) |
Zoom, or pan. |
20.5. Vocabulary entry
| Shortcut | Action |
|---|---|
Enter (one-line field) |
Save the entry. |
Ctrl/Cmd+Enter |
Save the entry, from any field. |