Web OpenCCG
This repository builds a small nginx-webserver and python wrapper around OpenCCG using the GUM-space ontology, ready to run inside a docker container.
After an initial docker-compose up, the service can be queried using a simple POST request, e.g. using curl:
$ curl --data "Take the cup." localhost/parse
{"version": "2.1.0", "application": "web-openccg", "uuid": "16ae93f7-8f61-4a5a-9ca7-70ec3dc27283", "sentence": "take the cup", "parses": {"smain": "@x1:gum-OrientationChange( <mood>imp ^ <gs-direction>(x2:gs-GeneralizedLocation ^ <gs-hasSpatialModality>(w2:slm-Cup ^ cup ^ <det>the ^ <ident>specific ^ <quant>singular)) ^ <gum-processInConfiguration>(w0:slm-Moving ^ slm-Taking))", "smain/0": "@x1:gum-OrientationChange( <mood>imp ^ <gs-direction>(x2:gs-GeneralizedLocation ^ <gs-hasSpatialModality>(w2:slm-Cup ^ cup ^ <det>the ^ <ident>specific ^ <quant>singular)) ^ <gum-processInConfiguration>(w0:slm-Taking ^ slm-Taking))", "smain/.r": "@x1:gs-AffectingDirectedMotion( <mood>imperative ^ <gs-route>x2 ^ <gum-actee>(w2:slm-Cup ^ cup ^ <det>the ^ <ident>specific ^ <quant>singular) ^ <gum-processInConfiguration>(w0:slm-Moving ^ slm-Taking))", "smain/.r/0": "@x1:gs-AffectingDirectedMotion( <mood>imperative ^ <gs-route>x2 ^ <gum-actee>(w2:slm-Cup ^ cup ^ <det>the ^ <ident>specific ^ <quant>singular) ^ <gum-processInConfiguration>(w0:slm-Taking ^ slm-Taking))"}, "http_status": 200, "json_parses": {"smain": {"__class__": "Nominal", "name": "x1", "roles": [{"__class__": "Role", "target": "imp", "type": "mood"}, {"__class__": "Role", "target": {"__class__": "Variable", "name": "x2", "roles": [{"__class__": "Role", "target": {"__class__": "Variable", "name": "w2", "roles": [{"__class__": "Role", "target": "cup", "type": "entity"}, {"__class__": "Role", "target": "the", "type": "det"}, {"__class__": "Role", "target": "specific", "type": "ident"}, {"__class__": "Role", "target": "singular", "type": "quant"}], "type": "slm-Cup"}, "type": "gs-hasSpatialModality"}], "type": "gs-GeneralizedLocation"}, "type": "gs-direction"}, {"__class__": "Role", "target": {"__class__": "Variable", "name": "w0", "roles": [{"__class__": "Role", "target": "slm-Taking", "type": "entity"}], "type": "slm-Moving"}, "type": "gum-processInConfiguration"}], "type": "gum-OrientationChange"}, "smain/0": {"__class__": "Nominal", "name": "x1", "roles": [{"__class__": "Role", "target": "imp", "type": "mood"}, {"__class__": "Role", "target": {"__class__": "Variable", "name": "x2", "roles": [{"__class__": "Role", "target": {"__class__": "Variable", "name": "w2", "roles": [{"__class__": "Role", "target": "cup", "type": "entity"}, {"__class__": "Role", "target": "the", "type": "det"}, {"__class__": "Role", "target": "specific", "type": "ident"}, {"__class__": "Role", "target": "singular", "type": "quant"}], "type": "slm-Cup"}, "type": "gs-hasSpatialModality"}], "type": "gs-GeneralizedLocation"}, "type": "gs-direction"}, {"__class__": "Role", "target": {"__class__": "Variable", "name": "w0", "roles": [{"__class__": "Role", "target": "slm-Taking", "type": "entity"}], "type": "slm-Taking"}, "type": "gum-processInConfiguration"}], "type": "gum-OrientationChange"}, "smain/.r": {"__class__": "Nominal", "name": "x1", "roles": [{"__class__": "Role", "target": "imperative", "type": "mood"}, {"__class__": "Role", "target": {"__class__": "Variable", "name": "x2", "roles": [{"__class__": "Role", "target": {"__class__": "Variable", "name": "w2", "roles": [{"__class__": "Role", "target": "cup", "type": "entity"}, {"__class__": "Role", "target": "the", "type": "det"}, {"__class__": "Role", "target": "specific", "type": "ident"}, {"__class__": "Role", "target": "singular", "type": "quant"}], "type": "slm-Cup"}, "type": "gum-actee"}, {"__class__": "Role", "target": {"__class__": "Variable", "name": "w0", "roles": [{"__class__": "Role", "target": "slm-Taking", "type": "entity"}], "type": "slm-Moving"}, "type": "gum-processInConfiguration"}], "type": null}, "type": "gs-route"}], "type": "gs-AffectingDirectedMotion"}, "smain/.r/0": {"__class__": "Nominal", "name": "x1", "roles": [{"__class__": "Role", "target": "imperative", "type": "mood"}, {"__class__": "Role", "target": {"__class__": "Variable", "name": "x2", "roles": [{"__class__": "Role", "target": {"__class__": "Variable", "name": "w2", "roles": [{"__class__": "Role", "target": "cup", "type": "entity"}, {"__class__": "Role", "target": "the", "type": "det"}, {"__class__": "Role", "target": "specific", "type": "ident"}, {"__class__": "Role", "target": "singular", "type": "quant"}], "type": "slm-Cup"}, "type": "gum-actee"}, {"__class__": "Role", "target": {"__class__": "Variable", "name": "w0", "roles": [{"__class__": "Role", "target": "slm-Taking", "type": "entity"}], "type": "slm-Taking"}, "type": "gum-processInConfiguration"}], "type": null}, "type": "gs-route"}], "type": "gs-AffectingDirectedMotion"}}}Or, as an example, using Python requests:
import requests
print(requests.post('http://localhost/parse', data={'sentence': 'Take the cup.'}).json())Note that is is not production ready, as it is really slow and not optimized: Instead of keeping one (or multiple) instances of OpenCCG running to query them faster, each request spawns an individual OpenCCG instance.
Usage
Querying
To query the service visually, just open your browser at http://localhost/. Otherwise, use curl, wget, or e.g. python requests to query web-openccg via the command line or your application.
If your client allows to build your request body manually, like curl, just put the sentence inside:
curl --data "Take the cup." localhost/parseHowever, many high level frameworks like python requests usually use a
key-value mechanism for post data. In this case, use the key sentence:
requests.post('http://localhost/parse', data={'sentence': 'Take the cup.'})Response format
The response is a JSON object and always contains these fields:
-
version: The JSON object version. -
application: Always "web-openccg", this is useful if you aggregate parses from different services. -
uuid: A unique ID for this response. This will only be useful if you plan to integrate the tool somehow. -
http_status: The HTTP status from the request.
If a sentence was provided during the request, these fields are present:
-
sentence: The cleaned input sentence (all lowercase, punctuation removed, ...).
If at least one successful parse exists, the these fields are included:
-
parses: A dictionary of parse-identifiers (e.g. "np") to actual parses as OpenCCG outputs them. -
json_parses: A version of the OpenCCG outputs in a flat JSON. This is produced via a custom grammar for TatSu. For details see below
Note: The keys are shared between parses and json_parses, thus you can easily lookup the original output for a JSON parse and vice-versa.
If an error occurs, the error field is present:
-
error: An error description.
An example response for the sentence "Take the cup." is:
{
"version": "2.1.0",
"application": "web-openccg",
"uuid": "14409a2a-d114-4c92-8cf9-56a509efd07f",
"sentence": "take the cup",
"parses": {
"smain": "@x1:gum-OrientationChange( <mood>imp ^ <gs-direction>(x2:gs-GeneralizedLocation ^ <gs-hasSpatialModality>(w2:slm-Cup ^ cup ^ <det>the ^ <ident>specific ^ <quant>singular)) ^ <gum-processInConfiguration>(w0:slm-Moving ^ slm-Taking))",
"smain/0": "@x1:gum-OrientationChange( <mood>imp ^ <gs-direction>(x2:gs-GeneralizedLocation ^ <gs-hasSpatialModality>(w2:slm-Cup ^ cup ^ <det>the ^ <ident>specific ^ <quant>singular)) ^ <gum-processInConfiguration>(w0:slm-Taking ^ slm-Taking))",
"smain/.r": "@x1:gs-AffectingDirectedMotion( <mood>imperative ^ <gs-route>x2 ^ <gum-actee>(w2:slm-Cup ^ cup ^ <det>the ^ <ident>specific ^ <quant>singular) ^ <gum-processInConfiguration>(w0:slm-Moving ^ slm-Taking))",
"smain/.r/0": "@x1:gs-AffectingDirectedMotion( <mood>imperative ^ <gs-route>x2 ^ <gum-actee>(w2:slm-Cup ^ cup ^ <det>the ^ <ident>specific ^ <quant>singular) ^ <gum-processInConfiguration>(w0:slm-Taking ^ slm-Taking))"
},
"http_status": 200,
"json_parses": {
"smain": {
"__class__": "Nominal",
"name": "x1",
"roles": [
{
"__class__": "Role",
"target": "imp",
"type": "mood"
},
{
"__class__": "Role",
"target": {
"__class__": "Variable",
"name": "x2",
"roles": [
{
"__class__": "Role",
"target": {
"__class__": "Variable",
"name": "w2",
"roles": [
{
"__class__": "Role",
"target": "cup",
"type": "entity"
},
{
"__class__": "Role",
"target": "the",
"type": "det"
},
{
"__class__": "Role",
"target": "specific",
"type": "ident"
},
{
"__class__": "Role",
"target": "singular",
"type": "quant"
}
],
"type": "slm-Cup"
},
"type": "gs-hasSpatialModality"
}
],
"type": "gs-GeneralizedLocation"
},
"type": "gs-direction"
},
{
"__class__": "Role",
"target": {
"__class__": "Variable",
"name": "w0",
"roles": [
{
"__class__": "Role",
"target": "slm-Taking",
"type": "entity"
}
],
"type": "slm-Moving"
},
"type": "gum-processInConfiguration"
}
],
"type": "gum-OrientationChange"
},
"smain/0": {
"__class__": "Nominal",
"name": "x1",
"roles": [
{
"__class__": "Role",
"target": "imp",
"type": "mood"
},
{
"__class__": "Role",
"target": {
"__class__": "Variable",
"name": "x2",
"roles": [
{
"__class__": "Role",
"target": {
"__class__": "Variable",
"name": "w2",
"roles": [
{
"__class__": "Role",
"target": "cup",
"type": "entity"
},
{
"__class__": "Role",
"target": "the",
"type": "det"
},
{
"__class__": "Role",
"target": "specific",
"type": "ident"
},
{
"__class__": "Role",
"target": "singular",
"type": "quant"
}
],
"type": "slm-Cup"
},
"type": "gs-hasSpatialModality"
}
],
"type": "gs-GeneralizedLocation"
},
"type": "gs-direction"
},
{
"__class__": "Role",
"target": {
"__class__": "Variable",
"name": "w0",
"roles": [
{
"__class__": "Role",
"target": "slm-Taking",
"type": "entity"
}
],
"type": "slm-Taking"
},
"type": "gum-processInConfiguration"
}
],
"type": "gum-OrientationChange"
},
"smain/.r": {
"__class__": "Nominal",
"name": "x1",
"roles": [
{
"__class__": "Role",
"target": "imperative",
"type": "mood"
},
{
"__class__": "Role",
"target": {
"__class__": "Variable",
"name": "x2",
"roles": [
{
"__class__": "Role",
"target": {
"__class__": "Variable",
"name": "w2",
"roles": [
{
"__class__": "Role",
"target": "cup",
"type": "entity"
},
{
"__class__": "Role",
"target": "the",
"type": "det"
},
{
"__class__": "Role",
"target": "specific",
"type": "ident"
},
{
"__class__": "Role",
"target": "singular",
"type": "quant"
}
],
"type": "slm-Cup"
},
"type": "gum-actee"
},
{
"__class__": "Role",
"target": {
"__class__": "Variable",
"name": "w0",
"roles": [
{
"__class__": "Role",
"target": "slm-Taking",
"type": "entity"
}
],
"type": "slm-Moving"
},
"type": "gum-processInConfiguration"
}
],
"type": null
},
"type": "gs-route"
}
],
"type": "gs-AffectingDirectedMotion"
},
"smain/.r/0": {
"__class__": "Nominal",
"name": "x1",
"roles": [
{
"__class__": "Role",
"target": "imperative",
"type": "mood"
},
{
"__class__": "Role",
"target": {
"__class__": "Variable",
"name": "x2",
"roles": [
{
"__class__": "Role",
"target": {
"__class__": "Variable",
"name": "w2",
"roles": [
{
"__class__": "Role",
"target": "cup",
"type": "entity"
},
{
"__class__": "Role",
"target": "the",
"type": "det"
},
{
"__class__": "Role",
"target": "specific",
"type": "ident"
},
{
"__class__": "Role",
"target": "singular",
"type": "quant"
}
],
"type": "slm-Cup"
},
"type": "gum-actee"
},
{
"__class__": "Role",
"target": {
"__class__": "Variable",
"name": "w0",
"roles": [
{
"__class__": "Role",
"target": "slm-Taking",
"type": "entity"
}
],
"type": "slm-Taking"
},
"type": "gum-processInConfiguration"
}
],
"type": null
},
"type": "gs-route"
}
],
"type": "gs-AffectingDirectedMotion"
}
}
}JSON format
Current version: 2.1.0
The JSON format for the OpenCCG parses can be determined from the example above or by carefully inspecting the fully typed OpenCCG.ebnf.
There are three different types of objects: Nominal, Variable, and Role. The full semantic specification (the JSON file) can either be one single entity of any of the three types or a list of Nominals. If we find parses which are lists of variables or roles, the grammar will be extended. Please open an issue if you find sentences which can not be parsed correctly.
Nominal and Variable
A nominal is like a special variable, it describes the "main theme" of the sentence. Since they are practically the same (they actually have very similar parsing rules), the JSON format represents them with an identical structure:
{
"__class__": "Variable",
"name": "x1",
"type": "gum-OrientationChange",
"roles": []
}-
__class__is eitherVariableorNominal. -
nameis the variable named used by OpenCCG, it is a letter followed by a number, for examplex1orw12. -
typeis a GUM specifier denoting the type of the variable. It can also benull, if the type is not specified. -
rolesis a list of roles as described below.
Role
A Role defines all properties a Variable or Nominal can have. It follows a very similar structure:
{
"__class__": "Role",
"type": "quant",
"target": ...
}-
__class__:Role. -
type: The role type. This determines thetargetpossibilities. -
target: The value of the role. This can be multiple different things, see below.
If the type is entity, the target will be an instance, like "cup" or "slm-Taking".
If the type has a prefix followed by a dash, e.g. gs-hasSpatialModality or gs-direction, then the target will be a Variable.
If the type is any other string, the target will be an atomic string, e.g. the type det can have the target the, while the type quant can have the target singular.
Changing the port
Most webservices use port 80 as a default port, and so does web-openccg.
To change the port, adjust the docker-compose file and change the port line
from "80:80" to your port on the left side (but keep the 80 on the right), so
for example to set up the service on Port 9043, you would change it to
"9043:80".
Development
Although not necessarily needed, I use a pipenv for local development to be able to compile the grammar to a parser using TatSu. I rely on when-changed to trigger automatic builds:
when-changed OpenCCG.ebnf makeTo start the development docker container, use the Makefile:
make runThe development server binds to port 5000 and uses the
flask debug environment. Additionally, the docker
container started with make run binds the app directory
so that flask's reloading works properly.