Glossary

Glossary#

Terms as they are used on the Swiss-AL Platform and in this documentation. Entries are alphabetical; use your browser’s find function or the index to locate a term quickly.

Absolute frequency#

The raw number of times something occurs in a corpus. Not comparable between corpora of different sizes — see relative frequency.

The search mode that accepts CQP queries, allowing you to combine annotations, match sequences of words, and use regular expressions. See Advanced Search.

Algorithm#

A systematic sequence of operations used to solve a problem or process data.

Annotation#

Information added to a text by automatic analysis rather than present in the original — lemma, part of speech, and named entity labels are all annotations. Added by the pipeline.

Background topic#

A topic that collects high-frequency, general-purpose words rather than representing a distinct theme. Most models produce one or two.

The search mode that matches words exactly as written and supports truncation, OR, and part of speech filtering. See Basic Search.

Category#

A grouping of corpus sources used to break down results. Which categories exist depends on the corpus type.

Coherence score#

A measure of how semantically similar a topic’s top words are to each other. Used to rank the documents most strongly associated with a topic.

Collocation#
Common pairing#
Co-occurrence#

Words that frequently occur together within a context window. Such recurring combinations can reveal conventional patterns of language use. On the platform, measured with LogDice. See Common Pairings.

Context window#

A span of n words before and after a search term, within which co-occurrence is counted. Five words either side is the default for both common pairings and word embeddings.

Corpus#

A structured collection of texts assembled for linguistic analysis. See Corpora.

Corpus keywords#

Words that occur considerably more frequently in a study corpus than in a reference corpus. Distinct from document keywords. See Keywords.

Corpus type#

The category a corpus belongs to — organizational, journalistic, parliamentary, project, or demo. The type determines how the corpus sources are subcategorized. See Corpus types.

Cosine similarity#

A measure of how similar two word vectors are, based on the angle between them rather than the distance between their endpoints. Higher values mean more similar contexts of use.

CQP#
Corpus Query Processor#

A query language for searching corpora using complex patterns and annotations, used by Advanced Search.

Demo corpus#

A subset of an organizational corpus, created for testing and demonstration. The only corpora accessible without registration. See Demo Corpora.

Distributional hypothesis#

The principle that words occurring in similar contexts tend to have similar meanings. The basis of semantic space methods.

Document keywords#

Words that occur more frequently in an individual document than in the corpus as a whole, adjusted for text length. Distinct from corpus keywords.

Field of activity#

A subcategory of organizational corpus sources, such as federation or trade association. Paired with social system.

Gibbs sampling#

The iterative procedure by which LDA reassigns words to topics until the assignments stabilize.

KWIC#
Key Word in Context#

A display format showing each occurrence of a search term with its immediate context, the search term aligned down the centre. See Context of Words.

LDA#
Latent Dirichlet Allocation#

The topic model used on the platform. A probabilistic model that treats documents as mixtures of topics and topics as distributions over words.

Lemma#

The base form of a word, under which it is listed in a dictionary — go for goes and went. Grouping by lemma collects all inflected forms together.

Lemmatization#

The automatic process of determining each token’s lemma.

Log likelihood ratio#
LLR#

A measure of how confident we can be that a word’s frequency difference between two corpora is not due to chance. Takes corpus size into account. Does not measure the size of the difference.

Log ratio#

A measure of the size of a frequency difference between two corpora, expressed as the binary logarithm of the ratio of relative frequencies. Each additional point represents a doubling.

LogDice#

The association measure used for common pairings. Runs to a theoretical maximum of 14, with each point representing a doubling of association strength. Independent of corpus size.

Media type#

A subcategory of journalistic corpus sources, such as daily/online newspaper or radio/tv. Paired with publisher.

Metadata#

Information about a text rather than its content — its date, source, title, and categories.

Moving average#

A smoothing technique that plots the average frequency over a rolling window of days, reducing daily fluctuation so longer trends become visible.

Named entity#

A token or sequence of tokens referring to a specific person, organization, place, or similar. Annotated automatically and searchable in Advanced Search.

Nearest neighbour#

The words closest to a given word in a semantic space, meaning those used in the most similar contexts. Not necessarily words with similar meanings.

OER#
Open Educational Resources#

Teaching and learning materials published under an open licence that permits reuse and adaptation. See Open Educational Resources.

ORD#
Open Research Data#

Scientific research data that is publicly accessible, to promote transparency and reuse.

Parent corpus#

The corpus a subcorpus was derived from. In a keyword analysis it serves as the reference corpus.

Part of speech#
PoS#

The grammatical category of a word, such as noun, verb, or adjective.

Pipeline#

The sequence of processing steps that extracts, filters, and annotates text before it enters a corpus. See Swiss-AL Pipeline.

Project#

A workspace container grouping your corpora and analyses. Creating one is required for access to the full corpus collection.

Publisher#

A subcategory of journalistic corpus sources, identifying who published a text. Paired with media type.

The simplest search mode: case-insensitive, matching all forms in a word’s paradigm. See Quick Search.

Reference corpus#

The corpus a study corpus is compared against in a keyword analysis. For a subcorpus, this is its parent corpus, with the overlapping documents excluded.

Regular expressions#
Regex#

A pattern-matching notation for describing sets of strings, available in Advanced Search.

Relative frequency#

A frequency count normalized by corpus size, allowing comparison between corpora of different sizes. Usually expressed in words per million.

Semantic space#
Vector space#

A mathematical representation of word meanings, in which words are positioned according to the contexts they occur in. Words used similarly sit closer together. See Semantic Space.

Social system#

A subcategory of organizational corpus sourcespolitics, industry, science, or civil society. Paired with field of activity.

Source#

A publication, website, or institution that texts in a corpus come from. Subcategorized according to the corpus type.

STTS#
Stuttgart-Tübingen Tagset#

The tagset used to annotate part of speech in the German corpora. See Tagsets.

Study corpus#

The corpus under investigation in a keyword analysis, compared against a reference corpus.

Subcorpus#

A corpus you build yourself by filtering a larger one by category, time span, source, or keyword. Required for corpus keywords. See Creating subcorpora.

t-SNE#

A dimensionality reduction technique used to draw high-dimensional data in two or three dimensions. Preserves local neighbourhoods better than global structure, so proximity is more meaningful than distance.

Tag#

A label attached to a token indicating its word class or grammatical properties.

Tagset#

The full inventory of tags used for a language. See Tagsets.

Token#

A single unit of a text — a word, number, or punctuation mark. Corpus size is measured in tokens.

Topic#

A cluster of words that frequently appear together across a set of documents and together suggest a common theme. See Topics.

Topic model#

A method for discovering topics in a collection of documents without being told in advance what those topics are.

Type#

A distinct word form in a corpus, as opposed to each of its occurrences. A text with 100 tokens may contain only 60 types.

Word embedding#

A word vector learned from a corpus by a model such as word2vec. See Semantic Space.

Word vector#

A numerical representation of a word in a semantic space, derived from the contexts the word occurs in.

word2vec#

The algorithm used on the platform to learn word embeddings, implemented via the Python library gensim.

Words per million#
WPM#

A relative frequency measure, standardized per 1,000,000 tokens, allowing comparison across corpora of different sizes.

Workspace#

The Swiss-AL interface: corpus and tool selection on the left, tabbed analysis area in the centre. See Quickstart Guide.