AI Glossary / Graphs and symbolic AI / Intermediate

Linguistic Linked Open Data

Definition

Linguistic Linked Open Data, or LLOD, is an approach and community movement for publishing and connecting language resources using linked-data principles and open licensing. It applies shared identifiers and machine-readable relationships to resources such as dictionaries, corpora, lexical networks, and linguistic annotations so that their information can be discovered and combined.

What it means

Language resources often describe related material in different formats and with different local identifiers. One dictionary may describe a word, another resource its meanings, and a corpus examples of its use. LLOD aims to make those connections explicit rather than leaving every user to reconstruct them.
Linked means that resources refer to one another through identifiable entities and stated relationships. Open concerns reuse permissions, not merely whether a page can be viewed in a browser. Linguistic identifies the subject matter. A collection of hyperlinks between ordinary dictionary pages is not automatically a well-modeled linguistic data graph.

How it works

Publishers identify entities using URIs, provide information through web-compatible representations, and add meaningful links to other resources. RDF is commonly used to express statements as subject-predicate-object triples. Suitable vocabularies distinguish entities such as entries, forms, senses, and annotations.
A consumer can follow identifiers or query available data to connect information across resources. Compatibility still requires careful modeling: two datasets using RDF do not automatically mean the same thing by their labels. Provenance, language metadata, licensing, and maintenance information help consumers assess what a link actually supports and whether the data is suitable for reuse.

Three practical examples

• A terminology service can connect a domain concept with language-specific lexical entries and definitions.
• A corpus project can link an annotated text span to an identified lexical sense while preserving the original annotation source.
• A multilingual search project can use curated relationships between language resources to expand retrieval beyond exact character-string matches.

Think of it this way

Think of a network of library catalogs that use reliable identifiers and carefully named relationships to point to the same books, editions, or subjects. The analogy is limited because linguistic entities are not interchangeable: a spelling, a word sense, and a concept may need separate identities even when people casually call all of them a word.

Worked example

Imagine a small fictional language-resource project with two entries for the English spelling bank. Entry E1 represents the financial-institution sense; E2 represents the river-edge sense. A corpus contains annotation A1 for bank in the sentence about opening an account, and annotation A2 for bank in a sentence about sitting beside a river.
The project records A1 as referring to E1's sense and A2 as referring to E2's sense. A separate concept resource provides C1 for a financial institution and C2 for a river edge. The corresponding senses are linked to the appropriate concepts using a relationship whose meaning is documented.
Suppose another dataset provides an entry in a different language linked to C1. A query following the sense-to-concept links can find that entry as a candidate for the financial context without also retrieving the river-edge entry merely because both English strings are bank. The exact relationship may still need a translator's review; a shared concept link does not establish interchangeable wording in every sentence.
Count the modeled distinctions: two English entries, two annotations, and two concepts remain six separately identified entities. Collapsing E1 and E2 because their written labels match would incorrectly connect both annotations to both concepts. Likewise, declaring every related entity identical would erase the difference between a text occurrence and a dictionary sense.
These names are teaching labels, not deployed URIs or claims about an existing dictionary. The example shows why explicit identity and relationship choices matter more than simply exporting rows into another file format.

Illustrative case study

In a fictional project, Maple Lexicon combines two public terminology resources by matching identical labels. A review finds that a technical abbreviation has different meanings in the two source domains. The resulting shared identifier incorrectly mixes definitions.
The team separates the entities, records a domain-qualified relationship only where justified, and retains source provenance. This invented case illustrates a semantic integration error. Shared formats helped move the data, but they did not decide which concepts or senses should be equated.

Common misconceptions and mistakes

Linked data is not automatically open data, and an accessible download does not by itself establish reuse permission. Conversely, openly licensed data may still be poorly linked or difficult to combine.
Do not use an identity relationship merely to mean related to. Matching labels, translations, broader concepts, and identical entities are different relationships. A URI also does not guarantee that a description remains available forever. Finally, an interconnected graph is not automatically a verified knowledge source: annotation mistakes, conflicting definitions, and outdated links can propagate unless consumers inspect their provenance and intended meaning.

Business use

For a business language project, identify the integration problem first: terminology alignment, multilingual discovery, corpus annotation, or another concrete need. Decide which entities require stable identifiers and which relationships the application will actually use.
Assess source coverage, update practices, language metadata, and reuse terms before depending on a dataset. Keep source-specific assertions distinguishable when combining resources, especially when they disagree. Evaluate a representative set of links manually and test the queries needed by the application. The benefit should be demonstrated through reliable retrieval or reuse, rather than inferred from the presence of a large graph diagram.

People also ask

Is LLOD one database?

No. It describes an approach and ecosystem of connected language resources, which may be maintained separately.

Does the same spelling imply the same entity?

No. Different senses or entries can share a written form.

Does publishing RDF make data open?

No. Machine-readable representation and permission to reuse are separate matters that must both be addressed.

Check your understanding

Why must the two bank senses remain distinct in the worked example?

Answer: They refer to different meanings and connect to different concepts and text annotations. Merging them solely because their labels match would create incorrect cross-resource connections.

Related terms

• Linked data
• RDF
• Ontology
• Lexical resource
• Word sense
• Interoperability
• URI
• corpus annotation
• provenance
• SPARQL
• open data

Learn next

Choose a small dictionary sample containing an ambiguous word. Identify the entries, written forms, senses, and example occurrences separately before drawing links. Then examine a published linguistic vocabulary and compare its relationship meanings with the distinctions required by your sample.

More in Graphs and symbolic AI

• Linear search - Linear search finds an item by checking entries one at a time until a match is found or the searchable collection is exhausted.
• Linear separator - A linear separator is a decision boundary defined by a weighted sum of features plus an offset.
• Linear temporal logic - Linear temporal logic, or LTL, is a formal language for describing how conditions should hold along a sequence of system states.
• Linkage constraints - Linkage constraints are restrictions imposed by the physical connections among the links and joints of a mechanism.
• Liquid event - A liquid event is an event type or temporal property that, when true throughout an interval, is also true throughout its subintervals.
• Literal - A literal, in logic, is an atomic formula or the negation of an atomic formula. P is a positive literal, and NOT P is a negative literal.

Sources

Linguistic Linked Open Data community, principles and resource guidance: https://linguistic-lod.org/index Christian Chiarcos and colleagues, Linguistic Linked Open Data: Introduction and Overview: https://aclanthology.org/W13-5501.pdf

Previous: Linearly separable Next: Linguistic value