LCP CLI module

Command-line tool for converting CONLLU files and uploading the corpus to LCP

Installation

Make sure you have python 3.11+ with pip installed in your local environment, then run

pip install lcpcli

Usage

Example:

lcpcli -i ~/conll_ext/ -o ~/upload/ -m upload -k $API_KEY -s $API_SECRET -p my_project

Help:

lcpcli --help

lcpcli takes a corpus of CoNLL-U (PLUS) files and imports it to a project created in an LCP instance, such as catchphrase.

Besides the standard token-level CoNLL-U fields (form, lemma, upos, xpos, feats, head, deprel, deps) one can also provide document- and sentence-level annotations using comment lines in the files (see the CoNLL-U Format section)

A more advanced functionality, lcpcli supports annotations aligned at the character level, such as named entities. See the Named Entities section for more information

CoNLL-U Format

The CoNLL-U format is documented at: https://universaldependencies.org/format.html

The LCP CLI converter will treat all the comments that start with # newdoc KEY = VALUE as document-level attributes. This means that if a CoNLL-U file contains the line # newdoc author = Jane Doe, then in LCP all the sentences from this file will be associated with a document whose meta attribute will contain author: 'Jane Doe'

All other comment lines following the format # key = value will add an entry to the meta attribute of the segment corresponding to the sentence below that line (ie not at the document level)

The key-value pairs in the misc column in a column line will go in the meta attribute of the corresponding token, with the exceptions of these key-value combinations:

• SpaceAfter=Yes vs. SpaceAfter=No controls whether the token will be represented with a trailing space character in the database
• start=n.m|end=o.p will align tokens, segments (sentences) and documents along a temporal axis, where n.m and o.p should be floating values in seconds

See below how to report all the attributes in the template .json file

CoNLL-U PLUS

CoNLL-U PLUS is an extension to the CoNLLU-U format documented at: https://universaldependencies.org/ext-format.html

If your files start with a comment line of the form # global.columns = ID FORM LEMMA UPOS XPOS FEATS HEAD DEPREL DEPS MISC, lcpcli will treat them as CoNLL-U PLUS files and process the columns according to the names you set in that line

Named Entities

Besides the nested token-, segment- and document-level entities, you can use lcpcli to define other character-aligned entities, such as named entities. To do so, you will need to prepare your corpus as CoNLL-U PLUS files which must define a dedicated column, e.g. NAMEDENTITY:

# global.columns = ID FORM LEMMA UPOS NAMEDENTITY

All the tokens belonging to the same named entity should report the same index in that column, or _ if it doesn't belong to a named entity. For example (assuming the columns defined above):

0	Christoph	Christoph	PROPN	114
1	W.	W.	PROPN	114
2	Bauer	Bauer	PROPN	114
3	erzählt	erzählen	VERB	_
4	die	der	DET	_
5	Geschichte	Geschichte	NOUN	_
6	von	von	ADP	_
7	Innsbruck	Innsbruck	PROPN	115
8	,	,	PUNCT	_

In this example, the three first tokens belong to the same named entity ("Christoph W. Bauer") and "Innsbruck" forms another named entity.

The directory containing your corpus files should also include one TSV file named after that column (in this example, it could be named namedentity.tsv -- the extension doesn't matter, as long as it's not the same as your CoNLLU files). Its first line should report headers, starting with ID and then any attributes associated with a named entity. The value in the first column for all the other lines should correspond to the ones listed in the CoNLLU file(s). For example:

ID	type
114	PER
115	LOC

Along with the CoNLLU-PLUS lines above, this would associate the corresponding occurrence of the sequence "Christoph W. Bauer" with a named entity of type PER and the corresponding occurrence of "Innsbruck" with a named entity of type LOC.

Finally, you need to report a corresponding entity type in the template .json under the layer key, for example:

"NamedEntity": {
    "abstract": false,
    "layerType": "span",
    "contains": "Token",
    "attributes": {
        "type": {
            "isGlobal": false,
            "type": "categorical",
            "values": [
                "PER",
                "LOC",
                "MISC",
                "ORG"
            ],
            "nullable": true
        }
    }
},

Make sure to set the abstract, layerType and contains attributes as illustrated above. See the section Convert and Upload for a full example of a template .json file.

Global attributes

In some cases, it makes sense for multiple entity types to share references: in those cases, they can define global attributes. An example of a global attribute is a speaker or an agent that can have a name, an age, etc. and be associated with both a segment (a sentence) and, say, a gesture. The corpus template would include definitions along these lines:

"globalAttributes": {
    "agent": {
        "type": "dict",
        "keys": {
            "name": {
                "type": "text"
            },
            "age": {
                "type": "number"
            }
        }
    }
},
"layer": {
    "Segment": {
        "abstract": false,
        "layerType": "span",
        "contains": "Token",
        "attributes": {
            "agent": {
                "ref": "agent"
            }
        }
    },
    "Gesture": {
        "abstract": false,
        "layerType": "unit",
        "anchoring": {
            "time": true
        },
        "attributes": {
            "agent": {
                "ref": "agent"
            }
        }
    }
}

You should include a file named global_attribute_agent.csv (mind the singular on attribute) with three columns: agent_id, name and age, and reference the values of agent_id appropriately as a sentence-level comment in your CoNLL-U files as well as in a file named gesture.tsv. For example:

global_attribute_agent.tsv:

agent_id	agent
10	{"name": "Jane Doe", "age": 37}

CoNLL-U file:

# sent_id = 1
# agent_id = 10
The the _ _ _

gesture.tsv:

gesture_id	agent_id	frame_range
1	10	[2345,5678]

Convert and Upload

1. Create a directory in which you have all your properly-fromatted CONLLU files

2. In the same directory, create a template .json file that describes your corpus structure (see above about the attributes key on Document and Segment), for example:

{
    "meta":{
        "name":"My corpus",
        "author":"Myself",
        "date":"2023",
        "version": 1,
        "corpusDescription":"This is my corpus"
    },
    "firstClass": {
        "document": "Document",
        "segment": "Segment",
        "token": "Token"
    },
    "layer": {
        "Token": {
            "abstract": false,
            "layerType": "unit",
            "anchoring": {
                "location": false,
                "stream": true,
                "time": false
            },
            "attributes": {
                "form": {
                    "isGlobal": false,
                    "type": "text",
                    "nullable": true
                },
                "lemma": {
                    "isGlobal": false,
                    "type": "text",
                    "nullable": false
                },
                "upos": {
                    "isGlobal": true,
                    "type": "categorical",
                    "nullable": true
                },
                "misc": {
                    "type": "jsonb"
                }
            }
        },
        "NamedEntity": {
            "abstract": false,
            "layerType": "span",
            "contains": "Token",
            "attributes": {
                "type": {
                    "isGlobal": false,
                    "type": "categorical",
                    "values": [
                        "PER",
                        "LOC",
                        "MISC",
                        "ORG"
                    ],
                    "nullable": true
                }
            }
        },
        "Segment": {
            "abstract": false,
            "layerType": "span",
            "contains": "Token"
        },
        "Document": {
            "abstract": false,
            "contains": "Segment",
            "layerType": "span",
            "attributes": {
                "meta": {
                    "Autor": {
                        "type": "text",
                        "nullable": true
                    },
                    "promethia_id": {
                        "type": "text",
                        "nullable": true
                    },
                    "ISBN": {
                        "type": "text",
                        "nullable": true
                    },
                    "Titel": {
                        "type": "text",
                        "nullable": true
                    }
                }
            }
        }
    }
}

3. If your corpus defines a character-anchored entity type such as named entities, make sure you also include a properly named and formatted TSV file for it in the directory (see the Named Entities section)

4. Visit an LCP instance (e.g. catchphrase) and create a new project if you don't already have one where your corpus should go

5. Retrieve the API key and secret for your project by clicking on the button that says: "Create API Key"

   The secret will appear at the bottom of the page and remain visible only for 120s, after which it will disappear forever (you would then need to revoke the API key and create a new one)

   The key itself is listed above the button that says "Revoke API key" (make sure to not copy the line that starts with "Secret Key" along with the API key itself)

6. Once you have your API key and secret, you can start converting and uploading your corpus by running the following command:

lcpcli -i $CONLLU_FOLDER -o $OUTPUT_FOLDER -m upload -k $API_KEY -s $API_SECRET -p $PROJECT_NAME --live

• $CONLLU_FOLDER should point to the folder that contains your CONLLU files
• $OUTPUT_FOLDER should point to another folder that will be used to store the converted files to be uploaded
• $API_KEY is the key you copied from your project on LCP (still visible when you visit the page)
• $API_SECRET is the secret you copied from your project on LCP (only visible upon API Key creation)
• $PROJECT_NAME is the name of the project exactly as displayed on LCP -- it is case-sensitive, and space characters should be escaped