Stream Corpus

YouTube Streaming Transcripts (2017-2025)

Corpus Overview

The Stream corpus contains transcripts from YouTube streaming channels, providing access to contemporary spoken discourse from online content creators.

The corpus has been processed using spaCy’s large model (en_core_web_lg) for linguistic annotation, including Named Entity Recognition (NER) and morphological features.

Key Statistics

Metric	Value
Total Words	~14 million
Time Period	2017-2025
Channels	3 (see below)
Language	Contemporary English (spoken)

Channels

Channel	Description
contrapoints	Philosophy, politics, culture
kelseydangerous	Gaming, commentary
moresimsie	Gaming, lifestyle

Available Attributes

The corpus includes the following searchable attributes:

Token Attributes

Attribute	Description	Example
`word`	Exact word form	[word="the"]
`lemma`	Base form (lemma)	[lemma="run"]
`tag`	POS tag (Penn Treebank)	[tag="NN.*"]
`head_n`	Head position (1-based, 0 for root)	Dependency parsing
`head`	Head lemma	Dependency parsing
`deprel`	Dependency relation	[deprel="nsubj"]
`ent_type`	Named entity type (ORG, PERSON, GPE, etc.)	[ent_type="PERSON"]
`ent_iob`	NER IOB tag (B=begin, I=inside, O=outside)	[ent_iob="B"]
`morph`	Morphological features (UD FEATS format)	[morph=".Tense=Past."]

Document Attributes

Attribute	Description	Example
`doc.id`	Document ID	Filter by specific video
`doc.channel`	Channel name	Filter: contrapoints, etc.
`doc.date`	Publication date	Filter by date range
`doc.year`	Publication year	Filter by year
`doc.title`	Video title	Reference metadata
`doc.url`	Video URL	Link to source
`doc.topic`	Topic tag (if available)	Topic filtering

CQL Query Examples

Basic Word Search

Find all occurrences of a word:

[lemma="stream"]

Named Entity Recognition

Find all person names:

[ent_type="PERSON"]

Find all organizations:

[ent_type="ORG"]

Find all locations (geopolitical entities):

[ent_type="GPE"]

Morphological Features

Find past tense verbs:

[morph=".Tense=Past."]

Find plural nouns:

[tag="NN.*" & morph=".Number=Plur."]

Find past tense of a specific verb:

[lemma="go" & morph=".Tense=Past."]

Dependency Relations

Find subjects of verbs:

[deprel="nsubj" & lemma="say"]

Combined Queries

Find past tense verbs followed by person names:

[morph=".Tense=Past."] []{0,3} [ent_type="PERSON"]

Filter by Channel

Search within a specific channel:

Use the metadata filter to select a channel (e.g., contrapoints)
Run: [lemma="philosophy"]{.cql}

Programmatic Access (API)

Query the corpus programmatically via the REST API:

Example Request (curl)

curl -X POST http://localhost:8000/cql/run \
  -H "Content-Type: application/json" \
  -d '{
    "cql": "[lemma=\"stream\"]",
    "corpus_id": "stream",
    "limit": 50
  }'

Quick Reference

Goal	CQL Query
Exact word	[word="stream"]
All forms (lemma)	[lemma="stream"]
Adjacent words	[lemma="live"] [lemma="stream"]
Person names	[ent_type="PERSON"]
Organizations	[ent_type="ORG"]
Past tense verbs	[morph=".Tense=Past."]
Plural nouns	[tag="NN.*" & morph=".Number=Plur."]