Home

Awesome

<p align="center"> <img src="https://github.com/annotation/stam/raw/master/logo.png" alt="stam logo" width="320" /> </p>

Crate Docs GitHub release Project Status: Active – The project has reached a stable, usable state and is being actively developed. Technology Readiness Level 7/9 - Release Candidate - Technology ready enough and in initial use by end-users in intended scholarly environments. Further validation in progress.

STAM Tools

A collection of command-line tools for working with STAM, a data-model for stand-off annotations on text.

Various tools are grouped under the stam tool, and invoked with a subcommand:

For many of these, you can set --verbose for extra details in the output.

Alternatively, the functionality provided by the tools is also exposed as a library via a Rust API.

Installation

From source

$ cargo install stam-tools

Demo

STAM demo

Usage

Add the --help flag after the subcommand for extensive usage instructions.

Most tools take as input a STAM JSON or CSV file containing an annotation store. You may also specify multiple stores which will be merged into one. Any files mentioned via the @include mechanism are loaded automatically.

When output is written, the first store file used as input is also used as output. You can prevent writing output files by setting --dry-run or prevent reusing the first input file by setting an explicit output using --output.

Instead of passing STAM JSON files, you can read from stdin and/or output to stdout by setting the filename to -, this works in many places.

These tools also support reading and writing STAM CSV.

Tools

stam init & stam annotate

The stam init command is used to initialize a new STAM annotationstore with resources (--resource, plain text or STAM JSON), annotation data sets (--annotationset, STAM JSON) and/or annotations (--annotations, JSON list of annotations in STAM JSON).

Example, the positional parameter (last one) is the annotation store to output, it may be STAM JSON or STAM CSV, determined by the file extension:

$ stam init --resource document.txt new.store.stam.json

The stam annotate command is almost identical to stam init, except it reads and modifies an existing annotation store, rather than starting a new one from scratch:

$ stam annotate --resource document.txt existing.store.stam.json

Whenever you load annotations and annotation data sets using these commands, they need to already be in STAM JSON format. To import data from other formats, use stam import instead.

The stam init and stam annotate commands are also capable of merging multiple annotation stores into one.

If you want to load a STAM annotationstore (or multiple) and save it under another name and/or other format, you can use stam init (or stam annotate) as well, they key is to then use an explicit --output filename that differs from the input. It serves to merge stores and/or convert between STAM JSON and STAM CSV. Example:

$ stam init --output merged.store.stam.csv mystore1.store.stam.json mystore2.store.stam.json

You can also pass STAMQL queries to stam annotate to add (or delete) annotations:

stam annotate --query 'ADD ANNOTATION WITH DATA "my-vocab" "type" "sentence"; TARGET ?x { SELECT TEXT ?x WHERE RESOURCE "smallquote.txt" OFFSET 0 25; }' demo.store.stam.json

stam info

The stam info command provides either some high-level details on the annotation store (number of resource, annotations, etc), or with the --verbose flag it goes as far as presenting, in a fairly raw format, all the data it holds.

Example:

$ stam info my.store.stam.json

stam query

The stam query tool is used to consult the annotation store and export selected STAM data into a simple tabular data format (TSV, tab separated values). You can configure precisely what columns you want to export using the --columns parameter, or simply rely on the defaults that are autodetected. See stam query --help for a list of supported columns.

A full query is done using the --query parameter and subsequently a query statement in the STAM Query Language (STAMQL):

Example 1) a query in STAMQL:

$ stam query --query 'SELECT ANNOTATION ?a WHERE DATA "myset" "pos" = "noun";'

However, if you simply want all annotations, resource, data, and don't want to formulate a query a shortcut is available by just the --type parameter to annotation,key,data,resource or dataset.

Example 2) get all annotations (also default behaviour if you omit --type and --query):

$ stam query --type annotation my.store.stam.json

For certain types, you can set --verbose to output more information, e.g. when querying for annotations it will also output all annotation data pertaining to the annotations. Do not that stam import can not import annotations back when you use this.

Example 3) get all annotations verbosely with all data:

$ stam query --verbose --type annotation my.store.stam.json

Example 4) get all keys:

$ stam query --type key my.store.stam.json

One of the more powerful functions is that you can specify custom columns by specifying a set ID, a delimiter and a key ID (the delimiter by default is a slash), for instance: my_set/part_of_speech. Such columns are automatically added for you if you have DATA or KEY constraints in your query (like in example 1), if that is not what you want, set --strict-columns. This custom column will hold the corresponding value if they key exists for the annotation.

Example 5) explicitly specified columns including a custom one:

$ stam query --columns Id,Text,TextResource,BeginOffset,EndOffset,my_set/part_of_speech my.store.stam.json

Example 6) Subqueries and multiple result variables

$ stam query --query 'SELECT ANNOTATION ?sentence WHERE DATA "myset" "type" = "sentence"; { SELECT ANNOTATION ?word WHERE RELATION ?sentence EMBEDS; DATA "myset" "type" = "word"; }'

This will result in a TSV file where the sentence will be repeated for each word that is found in it, a result number will be returned in a column, as well as the variable name.

The TSV output produced by this tool is not lossless, that is, it can not encode everything that STAM supports, unlike STAM JSON and STAM CSV. It does, however, give you a great deal of flexibility to quickly output only the data relevant for whatever your specific purpose is.

For queries that modify the annotation store, use stam annotate rather than stam query.

stam export

stam export is just an alias for stam query, their functionality is identical.

stam import

The stam import tool is used to import tabular data from a TSV (Tab Separated Values) file into STAM. Like stam query, you can configure precisely what columns you want to import, using the --columns parameter. By default, the import function will attempt to parse the first line of your TSV file as the header and use that to figure out the column configuration. You will often want to set --annotationset to set a default annotation set to use for custom columns. If you set --annotationset my_set then a column like part_of_speech will be interpreted in that set (same as if you wrote my_set/part_of_speech explicitly).

Here is a simple example of a possible import TSV file (with --annotationset my_set):

Text	TextResource	BeginOffset	EndOffset	part_of_speech
Hello	hello.txt	0	5	interjection
world	hello.txt	6	10	noun

The import function has some special abilities. If your TSV data does not mention specific offsets in a text resource(s), they will be looked up automatically during the import procedure! If the text resources don't even exist in the first place, they can be reconstructed (within certain constraints, the output text will likely be in tokenised form only). If your data does not explicitly reference a resource, use the --resource parameter to point to an existing resource that will act as a default, or --new-resource for the reconstruction behaviour.

By setting --resource hello.txt or --new-resource hello.txt you can import the following much more minimal TSV:

Text	part_of_speech
Hello	interjection
world	noun

The importer supports empty lines within the TSV file. When reconstructing text, these will map to (typically) a newline in the to-be-constructed text (this configurable with --outputdelimiter2). Likewise, the delimiter between rows is configurable with --outputdelimiter, and defaults to a space.

Note that stam import can not import everything stam query can export. It can only import rows exported with --type Annotation (the default), in which each row corresponds with one annotation.

stam grep

The stam grep tool can be used for matching regular expressions in text, it will return the resource identifiers, offsets and exact texts of all matching occurrences.

Example:

$ stam grep -e "[hzwHZW]ij" frogdeep.store.stam.json 
example.deep 690:693 Hij     1/1
example.deep    799:802 hij     1/1

The tab-separated columns are as follows:

  1. Resource ID
  2. Begin offset and end offset (non-inclusive) in unicode points
  3. The matching text
  4. The current capture group and total number of capture groups (if any)

stam tag

The stam tag tool can be used for matching regular expressions in text and subsequently associating annotations with the found results. It is a tool to do for example tokenization or other tagging tasks.

The stam tag command takes a TSV file (example) containing regular expression rules for the tagger. The file contains the following columns:

  1. The regular expressions follow the this syntax. The expression may contain one or or more capture groups containing the items that will be tagged, in that case anything else is considered context and will not be tagged.
  2. The ID of annotation data set
  3. The ID of the data key
  4. The value to set. If this follows the syntax $1,$2,etc.. it will assign the value of that capture group (1-indexed).

Example of the rules:

#EXPRESSION	#ANNOTATIONSET	#DATAKEY	#DATAVALUE
\w+(?:[-_]\w+)*	simpletokens	type	word
[\.\?,/]+	simpletokens	type	punctuation
[0-9]+(?:[,\.][0-9]+)	simpletokens	type	number

Example of applying this to a text resource:

# first we create a store and add a text resource
$ stam init --resource sometext.txt my.store.stam.json

# then we start the tagging
$ stam tag --rules rules.tsv my.store.stam.json

stam view

The stam view tool is used to visualize annotations. The default visualisation is HTML. This will output a self-contained static HTML document to standard output (the document does not reference any external assets). An alternative visualisation is text with ANSI escape codes for colours (--format ansi), which is suited for display in a terminal rather than a browser. The annotations you want to visualise are requested via queries in STAMQL, using the --query parameter.

The first query you have may contains subqueries which act as highlight queries. The main query is always the selection query, it determines what the main selection is and can be anything you can query that has text (i.e. resources, annotations, text selections).

Any subsequent queries are highlight queries, they determine what parts of the selections produced by the selection query you want to highlight. Highlighting is done by drawing a line underneath the text and optionally by a tag that shows extra information.

Instead of specifying subqueries, you may use the --query parameter multiple times to define subqueries via the command line. Always make sure these reference a variable defines in the main query.

STAM view example

Example with tags:

STAM view example with tags

Tags can be enabled by prepending the query/subquery (i.e. before SELECT) with one of the following attributes:

The first three tags use to the first DATA constraint found, if put before SELECT. Alternatively, they can also be directly put before (any) DATA constraint to explicitly select one.

Attributes may also be provided for styling HTML output, these go before the query/subquery as a whole:

If no attribute is provided, there will be no tags or styling shown for that query, only a highlight underline.

In the highlight queries, the variable from the main selection query is available and you should always use it in a constraint, otherwise performance will be sub-optimal! All your queries should have variable names and these will appear in the legend (unless you pass --no-legend).

Various real examples of visualisation and queries are shown here: https://github.com/knaw-huc/stam-experiments/tree/main/exp6

Example of ANSI output rather than HTML, using --format ansi:

STAM view example on the terminal with ANSI colours

stam align

The stam align tool is used to compute an alignment between two texts; it identifies which parts of the two texts are identical and computes a mapping between the two coordinate systems. Two related sequence alignments algorithms from bioinformatics are implemented to accomplish this: Smith-Waterman and Needleman-Wunsch. The score parameters to either are fully configurable.

The resulting alignment is added as an annotation, a so called transposition, according to the STAM Transpose extension.

This tool allows the alignment of any two text selections, which are passed via two --query parameters and take a query in STAMQL. Alternatively, if you want to align two resources (a common scenario), you can just use the --resource parameter, twice, as a more convenient shortcut.

Example invocation:

# first we create a store and add a two resource
$ stam init --resource text1.txt --resource text2.txt my.store.stam.json

# then we start the alignment (will be written to the annotation store)
$ stam align --verbose --resource text1.txt --resource text2.txt my.store.stam.json

With the --verbose flag, the alignment will be outputted to standard output in a simple TSV format with offsets for either sides, example excerpt:

/tmp/218.txt    1373-1439       /tmp/hoof001hwva02_01_0231.txt  1282-1348       "betoonen als dat van Weesp daer ick bij citatie in persoon tegens "   "betoonen als dat van Weesp daer ick bij citatie in persoon tegens "
/tmp/218.txt    1444-1508       /tmp/hoof001hwva02_01_0231.txt  1348-1412       "hem begost ende wijder voor heb te procederen tot alsulke peenen"     "hem begost ende wijder voor heb te procederen tot alsulke peenen"

You can also output transpositions and other alignments using the stam export --alignments (or stam query --alignments). This will output the same as above, except for an extra first column with the annotation (transposition) ID, and an extra final column with all annotations ID underlying the transposition (separated by a pipe character).

stam batch

The stam batch tool is used when you want to execute multiple subcommands in series.

Subcommands are read from standard input, either interactively or by piping input. The syntax for the subcommands is equivalent to their invocation from the command line, but with the following differences:

The annotation store(s) is loaded once at the start, and saved at the end if there are any changes (and you didn't set --dry-run). This gives stam batch its edge over just running the stam command itself in sequence; data need not be loaded and stored after each step.

stam fromxml

The stam fromxml tool allows to map XML files with inline annotations to STAM. It will effectively untangle the inline annotations and produce plain text on the one hand, and stand-off STAM annotations on that plain text on the other hand.

As there is an endless variety of XML formats imaginable, this tool takes as extra input an external configuration file that defines how to map from a specific XML format (e.g. xHTML, TEI or PageXML) to STAM. See for example this configuration for xHTML which contains some inline documentation to help you get going.

Example:

$ stam fromxml --inputfile tests/test.html --config config/fromxml/html.toml --force-new output.stam.json

Some notes:

stam split

With stam split you can split an annotation store by removing resources, data sets or annotations. The items to be removed are specified via the --query parameter (multiple allowed). The default behaviour is to remove the selected items, but you can reverse the behaviour by passing --keep; then all non-matching items will be removed. Use with --output to set an output filename if you don't want to overwrite and truncate your input store.

Instead of passing full queries, you can also pass resources or datasets directly via respectively --resource and --dataset.