zelph: A Sophisticated Semantic Network System

Introduction

zelph is an innovative semantic network system that allows inference rules to be defined within the network itself. This project provides a powerful foundation for knowledge representation and automated reasoning, with a special focus on efficiency and logical inference capabilities. With dedicated import functions and specialized semantic scripts (like wikidata.zph), zelph offers powerful analysis capabilities for the complete Wikidata knowledge graph while remaining adaptable for any semantic domain.

Quick Start Guide 🚀

see this page

Community and Support

Development of zelph is supported by the Wikimedia Community Fund.

The project addresses real-world challenges in large-scale ontology management through direct collaboration with the Wikidata Ontology Cleaning Task Force and the Mereology Task Force.

Components

The zelph ecosystem includes:

• A core C++ library providing both C++ and C interfaces
• A single command-line binary that offers both interactive usage (CLI) and batch processing capabilities
• API functions beyond what’s available in the command-line interface
• Integration options for languages like Go and Lua through the C interface

The key features of zelph include:

• Representation of knowledge in a semantic network structure
• Rules encoded within the same semantic network as facts
• Support for multi-language node naming
• Contradiction detection and resolution
• Memory-efficient data structures optimized at bit level
• A flexible scripting language for knowledge definition and querying
• Built-in import functionality for Wikidata JSON datasets

Core Concepts

Semantic Network Structure

In zelph, knowledge is represented as a network of nodes connected by relations. Unlike traditional semantic networks where relations are labeled edges, zelph treats relation types as first-class nodes themselves. This unique approach enables powerful meta-reasoning about relations.

Facts and Relations

Facts in zelph are represented as triples consisting of a subject, relation type, and object. The standard relation type is ~, which represents a categorical relation (similar to "is a" or "instance of"). For example:

X ~ Y

This means "X is an instance of category Y" or "X is a Y".

Working with Custom Relations

zelph can work with any type of relation, not just the standard ~ relation. Here’s how custom relations work:

is opposite of ~ ->
> «is opposite of» «~» «->»
white is opposite of black
> «white» «is opposite of» «black»

In this example, using the interactive CLI, the first line declares "is opposite of" as a relation type (a member of the -> category). After the > symbol, we see zelph’s responses.

zelph creates new relation types automatically as needed. The explicit declaration of "is opposite of" can actually be omitted:

white "is opposite of" black
> «white» «is opposite of» «black»

Here, zelph automatically recognizes that "is opposite of" must be a relation type. Note that when a relation contains spaces and is being used for the first time, it must be enclosed in quotation marks. Once the relation is known to zelph, the quotation marks are no longer necessary in subsequent usage.

Internal Representation of facts

In a conventional semantic network, relations between nodes are labeled, e.g.

graph LR
    white -->|is opposite of| black

zelph’s representation of relation types works fundamentally differently. As mentioned in the introduction, one of zelph’s distinguishing features is that it treats relation types as first-class nodes rather than as mere edge labels.

At the network level, there is only a single primitive relation type: ~, which represents a general category relation. Far from being a limitation, this is actually one of zelph’s most powerful characteristics. Relations that differ from the basic ~ type are not represented as arrow labels in zelph, but as regular nodes with the same status as any other node in the network.

Internally, zelph creates special nodes to represent relations. For example, when defining:

"is opposite of" ~ ->

This tells zelph that "is opposite of" is a relation (represented by ->, which is the category of all relations). zelph creates a special node to represent this fact.

This can be visualized as follows:

graph TD
    A("is opposite of ~ ->") <--> B("is opposite of")
    C("->") --> A
    A --> D("~")

The nodes -> and ~ are predefined zelph nodes. -> represents the category of all relations, while ~ represents a subset of this category, namely the category of categorical relations. Every relation that differs from the standard relation ~ (like "is opposite of") is linked to -> via a ~ relation.

The node is opposite of ~ -> represents this specific relation (hence its name). The relations to other nodes encode its meaning.

This approach provides several advantages:

1. It enables meta-reasoning about relations themselves
2. It simplifies the underlying data structures
3. It allows relations to participate in other relations (higher-order relations)
4. It provides a unified representation mechanism for both facts and rules

This architecture is particularly valuable when working with knowledge bases like Wikidata, where relations (called "properties" in Wikidata terminology) are themselves first-class entities with their own attributes, constraints, and relationships to other entities. zelph’s approach naturally aligns with Wikidata’s conceptual model, allowing for seamless representation and inference across the entire knowledge graph.

Similarly, when stating:

white "is opposite of" black

zelph creates a special relation node that connects the subject "white" bidirectionally, the object "black" in reverse direction, and the relation type "is opposite of" in the forward direction.

graph TD
    A("white is opposite of black") <--> B("white")
    C("black") --> A
    A --> D("is opposite of")

The directions of the relations are as follows:

Element	Example	Relation Direction
Subject	white	bidirectional
Object	black	backward
Relation Type	is opposite of	forward

This semantics is used by zelph in several contexts, such as rule unification. It’s required because zelph doesn’t encode relation types as labels on arrows but rather as equal nodes. This has the advantage of facilitating statements about statements, for example, the statement that a relation is transitive.

This design prevents subject and object from being identical in a relation. There are examples of this in Wikidata, e.g., "South Africa (Q258)" "country (P17)" "South Africa (Q258)". "South Africa" is thus linked to itself in Wikidata via the relation (property) "Country". These examples are extremely rare in Wikidata and are ignored during import, with a warning.

Creating a node graph

You can generate a node graph yourself using zelph’s .dot command, which outputs a GraphViz DOT format file. For example:

.dot name 2

In this example, name refers to the node identifier (in the currently active language specified via the .lang command) whose connections you want to visualize. The following number represents the depth of connections to include in the graph.

To convert the DOT file into an actual image, use the GraphViz command-line tool as follows:

dot -Tsvg -oname.svg name.dot

Rules and Inference

One of zelph’s most powerful features is the ability to define inference rules within the same network as facts. Rules are statements containing => with conditions before it and a consequence after it.

Example rule:

R ~ "transitive relation", X R Y, Y R Z => X R Z

This rule states that if R is a transitive relation, and X is related to Y by R, and Y is related to Z by R, then X is also related to Z by R. Variables in the zelph syntax are currently single uppercase letters. This restricts the number of variables in a rule to 26. Note that the scope of variables always is a single rule. Internally, more complex rules are possible, but this is currently only supported via zelph’s API, not via the scripting interface. See main.cpp for an example on how to use the API.

Here is a practical example of how this rule works in zelph (which you can also try out in interactive mode):

R ~ transitive relation, X R Y, Y R Z => X R Z
> ((Y R Z), (X R Y), (R «~» «transitive relation»)) «=>» (X R Z)

After the > symbol, we see zelph’s output, which in this case simply confirms the input of the rule. The brackets () indicate that their content is represented as a separate node - each condition is a separate node in the semantic network.

Now, let’s declare that the relation > (greater than) belongs to the category (~) of transitive relations:

> ~ transitive relation
> «>» «~» «transitive relation»

Next, we provide three elements ("4", "5" and "6") for which the > relation applies:

6 > 5
> «6» «>» «5»
5 > 4
> «5» «>» «4»
«6» «>» «4» ⇐ («5» «>» «4»), («6» «>» «5»), («>» «~» «transitive relation»)

After entering 5 > 4, zelph’s unification mechanism takes effect and automatically adds a new fact: 6 > 4. This demonstrates the power of the transitive relation rule in action.

Rules can also define contradictions using !:

X "is opposite of" Y, A ~ X, A ~ Y => !

This rule states that if X is opposite of Y, then an entity A cannot be both an X and a Y, as this would be a contradiction.

If a contradiction is detected when a fact is entered (via the scripting language or during import of Wikidata data), the corresponding relation (the fact) is not entered into the semantic network. Instead, a fact is entered that describes this contradiction (making it visible in the Markdown export of the facts).

Performing Inference

Facts and rules are added immediately, but inferences are only performed when you explicitly run .run.
Queries containing variables (e.g., A "is capital of" Germany) are answered immediately without .run.

After entering facts and rules (interactively or via script), start the inference engine with:

.run

This performs full inference: rules are applied repeatedly until no new facts can be derived. New deductions are printed as they are found.

For a single inference pass:

.run-once

To export all deductions and contradictions as structured Markdown reports:

.run-md <subdir>

This command generates a tree of Markdown files in mkdocs/docs/<subdir>/ (the directory mkdocs/docs/ must already exist in the current working directory).
It is intended for integrating detailed reports into an existing MkDocs site – this is exactly how the contradiction and deduction reports on https://zelph.org were produced.
For normal interactive or script use, .run is the standard command.

Exporting Deduced Facts to File

The command .run-file <path> performs full inference (like .run) but additionally writes every deduced fact (positive deductions and contradictions) to the specified file – one per line.

Key characteristics of the file output:

• Reversed order: The reasoning chain comes first, followed by ⇒ and then the conclusion (or ! for contradictions).
• Clean format: No «» markup, no parentheses, no additional explanations – only the pure facts.
• Console output unchanged: On the terminal you still see the normal format with ⇐ explanations and markup.

Example session (with .lang wikidata active):

> Q1 P1 Q2
«Q1» «P1» «Q2»
> Q2 P1 Q3
«Q2» «P1» «Q3»
> A P1 B, B P1 C => A P2 C
((A «P1» B), (B «P1» C)) «=>» (A «P2» C)
> .run-file /home/stefan/RAMDisk/output2.txt
Starting full inference in encode mode – deduced facts (reversed order, no brackets/markup) will be written to /home/stefan/RAMDisk/output2.txt (with Wikidata token encoding).
...
«Q1» «P2» «Q3» ⇐ («Q1» «P1» «Q2»), («Q2» «P1» «Q3»)
...
> Ready.

Content of output2.txt:

丂 一丂 七, 七 一丂 丄 ⇒ 丂 一七 丄

Decoding the file:

> .decode /home/stefan/RAMDisk/output2.txt
Q1 P1 Q2, Q2 P1 Q3 ⇒ Q1 P2 Q3

The command is general-purpose and works with any language setting. It simply collects all deductions in a clean, machine-readable text file.

When the current language is set to wikidata (via .lang wikidata), the output is automatically compressed using a dense encoding that maps Q/P identifiers to CJK characters. This dramatically reduces file size and – crucially – makes the data highly suitable for training or prompting large language models (LLMs). Standard tokenizers struggle with long numeric identifiers (Q123456789), often splitting them into many sub-tokens. The compact CJK encoding avoids this problem entirely, enabling efficient fine-tuning or continuation tasks on Wikidata-derived logical data.

To read an encoded (or plain) file back in human-readable form, use:

.decode <path>

This prints each line decoded (if it was encoded) using Wikidata identifiers.

Internal representation of rules

Let’s explain the internal representation of rules based on the example rule above. The complete rule graph looks like this:

graph TD
    n1["((Y R Z), (X R Y), (R «~» «transitive relation»)) «=>» (X R Z)"] --> n2["=>"]
    n3["->"] --> n4["R «~» «->»"]
    n10["transitive relation"] --> n11["R «~» «transitive relation»"]
    n12["(Y R Z), (X R Y), (R «~» «transitive relation»)"] <--> n1
    n12 --> n13[","]
    n4 --> n8["~"]
    n4 <--> n14["R"]
    n11 --> n8
    n11 --> n12
    n11 <--> n14
    n15["X R Y"] --> n12
    n15 --> n14
    n16["X R Z"] --> n1
    n16 --> n14
    n17["X"] <--> n15
    n17 <--> n16
    n18["Y R Z"] --> n12
    n18 --> n14
    n19["Y"] --> n15
    n19 <--> n18
    n20["Z"] --> n16
    n20 --> n18

    style n12 fill:#87CEFA
    style n14 fill:#EEE8AA
    style n16 fill:#B3EE3A

This graph may seem somewhat overwhelming at first glance, but it follows a clear structure. Let’s break it down:

1. The three conditions of the rule are connected to the blue condition node, which itself points to the logical operation of the condition: , (which represents the logical AND operation):

   graph TD
       n12["(Y R Z), (X R Y), (R «~» «transitive relation»)"] --> n13[","]
       n11["R «~» «transitive relation»"] --> n12
       n15["X R Y"] --> n12
       n18["Y R Z"] --> n12
   
       style n12 fill:#87CEFA

2. The blue condition node serves as the subject of the rule clause S => O (which is assigned the complete rule statement as a name). The green conclusion node functions as the object of the rule clause:

   graph TD
       n1["((Y R Z), (X R Y), (R «~» «transitive relation»)) «=>» (X R Z)"] --> n2["=>"]
       n12["(Y R Z), (X R Y), (R «~» «transitive relation»)"] <--> n1
       n16["X R Z"] --> n1
   
       style n12 fill:#87CEFA
       style n16 fill:#B3EE3A

3. Each condition, as well as the conclusion, is represented exactly like a fact (see the previous section "Internal Representation of facts").

This summarizes the complete diagram shown above. As mentioned earlier, the elegant aspect of this representation method is that the inference system can be applied not only to facts but also to rules. Consequently, it becomes possible to formulate rules that generate other rules.

Facts and Rules in One Network: Unique Identification via Topological Semantics

A distinctive aspect of zelph is that facts and rules live in the same semantic network. That raises a natural question: how does the unification engine avoid confusing ordinary entities with statement nodes, and how does it keep rule matching unambiguous?

The answer lies in the network’s strict topological semantics (see Internal Representation of facts and Internal representation of rules). In zelph, a statement node is not “just a node with a long label”; it has a unique structural signature:

• Bidirectional connection to its subject
• Forward connection to its relation type (a first-class node)
• Backward connection to its object

The unification engine is hard-wired to search only for this pattern when matching a rule’s conditions. In other words, a variable that ranges over “statements” can only unify with nodes that expose exactly this subject/rel/type/object wiring. Conversely, variables intended to stand for ordinary entities cannot accidentally match a statement node, because ordinary entities lack that tri-partite signature.

Two immediate consequences follow:

1. Unambiguous matching. The matcher cannot mistake an entity for a statement or vice versa; they occupy disjoint topological roles.
2. Network stability. Because statementhood is encoded structurally, rules cannot “drift” into unintended parts of the graph. This design prevents spurious matches and the sort of runaway growth that would result if arbitrary nodes could pose as statements.

These constraints are not merely aesthetic; they are core to zelph’s reasoning guarantees and underpin the termination argument below.

Example Script

Here’s a comprehensive example demonstrating zelph’s capabilities:

X "is a" Y  => X ~ Y
X "is an" Y => X "is a" Y

is               "is a" ->
"has part"       "is a" ->
"is opposite of" "is a" ->

"is attribute of" "is opposite of" is
"is part of"      "is opposite of" "has part"
"is for example"  "is opposite of" "is a"

"has part"      "is transitive"
"has attribute" "is transitive"
~               "is transitive"

R "is transitive", X R Y, Y R Z => X R Z
X is E, E "is a" K  => X is K
X "has part" P, P "is a" K  => X "has part" K
K is E, X "is a" K  => X is E
K "has part" P, X "is a" K  => X "has part" P
X "is opposite of" Y, X "is a" K => Y "is a" K
X "is opposite of" Y => Y "is opposite of" X
R "is opposite of" S, X R Y => Y S X

X "is opposite of" Y, A is X, A is Y => !
X "is opposite of" Y, A "has part" X, A "has part" Y => !
X "is opposite of" Y, A "is a" X, A "is a" Y => !
X is E, X "is a" E => !
X is E, E "is a" X => !
X is E, E "has part" X => !

generates "is a" ->
needs "is a" ->

"is needed by" "is opposite of" needs
"is generated by" "is opposite of" generates

X generates energy => X "is an" "energy source"
A is hot => A generates heat
A generates oxygen => A is alive

chimpanzee "is an" ape
ape is alive

chimpanzee "has part" hand
hand "has part" finger

"green mint" "is an" mint
"water mint" "is a" mint
peppermint "is an" mint
mint "is a" lamiacea
catnip "is a" lamiacea
"green mint" is sweet

"is ancestor of" "is transitive"
peter "is ancestor of" paul
paul "is ancestor of" pius
A "is ancestor of" pius

When executed, the last line is interpreted as a query, because it contains a variable (single uppercase letter) and is no rule. Here are the results:

Answer: «paul» «is ancestor of» «pius»
«catnip» «~» «lamiacea» ⇐ «catnip» «is a» «lamiacea»
«needs» «~» «->» ⇐ «needs» «is a» «->»
«water mint» «~» «mint» ⇐ «water mint» «is a» «mint»
«mint» «~» «lamiacea» ⇐ «mint» «is a» «lamiacea»
«chimpanzee» «has part» «finger» ⇐ («hand» «has part» «finger»), («chimpanzee» «has part» «hand»), («has part» «is» «transitive»)
«peter» «is ancestor of» «pius» ⇐ («paul» «is ancestor of» «pius»), («peter» «is ancestor of» «paul»), («is ancestor of» «is» «transitive»)
«water mint» «~» «lamiacea» ⇐ («mint» «~» «lamiacea»), («water mint» «~» «mint»), («~» «is» «transitive»)
«peppermint» «is a» «mint» ⇐ «peppermint» «is an» «mint»
«chimpanzee» «is a» «ape» ⇐ «chimpanzee» «is an» «ape»
«green mint» «is a» «mint» ⇐ «green mint» «is an» «mint»
«chimpanzee» «is» «alive» ⇐ («chimpanzee» «is a» «ape»), («ape» «is» «alive»)
«generates» «is opposite of» «is generated by» ⇐ «is generated by» «is opposite of» «generates»
«has part» «is opposite of» «is part of» ⇐ «is part of» «is opposite of» «has part»
«is a» «is opposite of» «is for example» ⇐ «is for example» «is opposite of» «is a»
«is» «is opposite of» «is attribute of» ⇐ «is attribute of» «is opposite of» «is»
«needs» «is opposite of» «is needed by» ⇐ «is needed by» «is opposite of» «needs»
«finger» «is part of» «hand» ⇐ («hand» «has part» «finger»), («has part» «is opposite of» «is part of»)
«hand» «is part of» «chimpanzee» ⇐ («chimpanzee» «has part» «hand»), («has part» «is opposite of» «is part of»)
«finger» «is part of» «chimpanzee» ⇐ («chimpanzee» «has part» «finger»), («has part» «is opposite of» «is part of»)
«sweet» «is attribute of» «green mint» ⇐ («green mint» «is» «sweet»), («is» «is opposite of» «is attribute of»)
«alive» «is attribute of» «ape» ⇐ («ape» «is» «alive»), («is» «is opposite of» «is attribute of»)
«transitive» «is attribute of» «is ancestor of» ⇐ («is ancestor of» «is» «transitive»), («is» «is opposite of» «is attribute of»)
«alive» «is attribute of» «chimpanzee» ⇐ («chimpanzee» «is» «alive»), («is» «is opposite of» «is attribute of»)
«transitive» «is attribute of» «has part» ⇐ («has part» «is» «transitive»), («is» «is opposite of» «is attribute of»)
«transitive» «is attribute of» «~» ⇐ («~» «is» «transitive»), («is» «is opposite of» «is attribute of»)
«transitive» «is attribute of» «has attribute» ⇐ («has attribute» «is» «transitive»), («is» «is opposite of» «is attribute of»)
«mint» «is for example» «green mint» ⇐ («green mint» «is a» «mint»), («is a» «is opposite of» «is for example»)
«lamiacea» «is for example» «catnip» ⇐ («catnip» «is a» «lamiacea»), («is a» «is opposite of» «is for example»)
«->» «is for example» «needs» ⇐ («needs» «is a» «->»), («is a» «is opposite of» «is for example»)
«mint» «is for example» «water mint» ⇐ («water mint» «is a» «mint»), («is a» «is opposite of» «is for example»)
«->» «is for example» «is» ⇐ («is» «is a» «->»), («is a» «is opposite of» «is for example»)
«->» «is for example» «has part» ⇐ («has part» «is a» «->»), («is a» «is opposite of» «is for example»)
«ape» «is for example» «chimpanzee» ⇐ («chimpanzee» «is a» «ape»), («is a» «is opposite of» «is for example»)
«lamiacea» «is for example» «mint» ⇐ («mint» «is a» «lamiacea»), («is a» «is opposite of» «is for example»)
«->» «is for example» «is opposite of» ⇐ («is opposite of» «is a» «->»), («is a» «is opposite of» «is for example»)
«->» «is for example» «generates» ⇐ («generates» «is a» «->»), («is a» «is opposite of» «is for example»)
«mint» «is for example» «peppermint» ⇐ («peppermint» «is a» «mint»), («is a» «is opposite of» «is for example»)
«green mint» «~» «mint» ⇐ «green mint» «is a» «mint»
«chimpanzee» «~» «ape» ⇐ «chimpanzee» «is a» «ape»
«peppermint» «~» «mint» ⇐ «peppermint» «is a» «mint»
«peppermint» «~» «lamiacea» ⇐ («mint» «~» «lamiacea»), («peppermint» «~» «mint»), («~» «is» «transitive»)
«green mint» «~» «lamiacea» ⇐ («mint» «~» «lamiacea»), («green mint» «~» «mint»), («~» «is» «transitive»)
«is needed by» «is a» «->» ⇐ («needs» «is a» «->»), («needs» «is opposite of» «is needed by»)
«is attribute of» «is a» «->» ⇐ («is» «is a» «->»), («is» «is opposite of» «is attribute of»)
«is part of» «is a» «->» ⇐ («has part» «is a» «->»), («has part» «is opposite of» «is part of»)
«is generated by» «is a» «->» ⇐ («generates» «is a» «->»), («generates» «is opposite of» «is generated by»)
«->» «is for example» «is generated by» ⇐ («is generated by» «is a» «->»), («is a» «is opposite of» «is for example»)
«->» «is for example» «is attribute of» ⇐ («is attribute of» «is a» «->»), («is a» «is opposite of» «is for example»)
«->» «is for example» «is needed by» ⇐ («is needed by» «is a» «->»), («is a» «is opposite of» «is for example»)
«->» «is for example» «is part of» ⇐ («is part of» «is a» «->»), («is a» «is opposite of» «is for example»)
«is generated by» «~» «->» ⇐ «is generated by» «is a» «->»
«is needed by» «~» «->» ⇐ «is needed by» «is a» «->»
Ready.

The results demonstrate zelph’s powerful inference capabilities. It not only answers the specific query about who is an ancestor of pius, but also derives numerous other facts based on the rules and base facts provided in the script.

Multi-language Support

zelph allows nodes to have names in multiple languages. This feature is particularly useful when integrating with external knowledge bases. The preferred language can be set in scripts using the .lang command:

.lang zelph

This capability is fully utilized in the Wikidata integration, where node names include both human-readable labels and Wikidata identifiers. An item in zelph can be assigned names in any number of languages, with Wikidata IDs being handled as a specific language ("wikidata").

Project Status

The project is currently in Version 0.9.2 (Beta). Core functionality is operational and has been rigorously tested against the full Wikidata dataset.

Current focus areas include:

• REPL and parser refinement: The REPL interface and the zelph language parser require architectural improvements.
• Enhancement of semantic rules: The wikidata.zph script serves as a base, but the strategy has shifted from generic deductions to targeted contradiction detection. See the Grant Report for details on this approach.
• Potential Wikidata integration: Exploring pathways for integration with the Wikidata ecosystem, e.g. the WikiProject Ontology.

Regarding potential Wikidata integration and the enhancement of semantic scripts, collaboration with domain experts would be particularly valuable. Expert input on conceptual alignment and implementation of best practices would significantly accelerate development and ensure optimal compatibility with existing Wikidata infrastructure and standards.

Building zelph

You need:

• C++ compiler (supporting at least C++20)
• CMake 3.25.2+
• Git

Build Instructions

1. Clone the repository with all submodules:

git clone --recurse-submodules https://github.com/acrion/zelph.git

1. Configure the build (Release mode):

cmake -D CMAKE_BUILD_TYPE=Release -B build src

1. Build the project (for MSVC, add --config Release):

cmake --build build

Verifying the Build

Test your installation by running the CLI:

./build/bin/zelph

or

./build/bin/zelph sample_scripts/english.zph

Licensing

zelph is dual-licensed:

1. AGPL v3 or later for open-source use,
2. Commercial licensing for closed-source integration or special requirements.

We would like to emphasize that offering a dual license does not restrict users of the normal open-source license (including commercial users). The dual licensing model is designed to support both open-source collaboration and commercial integration needs. For commercial licensing inquiries, please contact us at https://acrion.ch/sales.