AI Glossary / Natural language and speech / Advanced
PATR-II
Definition
PATR-II is a declarative formalism for representing linguistic information in computer-processable grammars. It combines context-free phrase-structure rules with feature structures and constraints enforced through unification. Developed in the early 1980s, it provided a way to express grammatical structure and information such as agreement without embedding every linguistic decision directly in a parser's procedural code.
What it means
The formalism describes what information must fit together for a grammatical analysis to be licensed. A phrase-structure rule supplies a structural pattern, while feature constraints express additional requirements among its constituents.
PATR-II is a grammar representation framework, not a complete grammar of English or a single theory explaining all languages. The analyst must supply rules, lexical entries, and appropriate features. A particular implementation also supplies parsing machinery and may add extensions. PC-PATR, for example, documents itself as an implementation of PATR-II with enhancements, so its details should not automatically be attributed to the original core formalism.
How it works
Feature structures associate labels with values. A value may be an atomic symbol or a further structure, and parts of a structure can be shared. The original account represents these descriptions using directed acyclic graphs rather than limiting them to independent, flat lists of attributes.
Unification combines compatible information. If two descriptions specify the same feature consistently, they can be combined. If one leaves a feature unspecified and the other supplies it, the combination can retain that information. Conflicting atomic requirements, such as singular and plural at the same required location, prevent successful unification.
A grammar rule connects these descriptions across a phrase's constituents. A subject and verb phrase can be required to share agreement information, for example. The parser seeks analyses satisfying both the structural rules and the feature constraints. A failed analysis can therefore reflect a structural mismatch, a feature conflict, or missing lexical coverage; these are different problems for the grammar author to investigate.
Three practical examples
1. Grammar engineering: A linguist expresses agreement requirements through shared features rather than multiplying category names for every combination of grammatical properties.
2. Teaching computational syntax: Students compare a phrase-structure-only grammar with a feature-constrained version to see which extra distinctions the constraints enforce.
3. Maintaining specialist language tools: A team inspecting an existing unification-based grammar traces why a particular sentence is admitted, rejected, or assigned multiple analyses.
Think of it this way
Think of joining two forms that must describe the same booking. One may supply a date and another a room, allowing the information to be combined. If both specify different dates for the same required field, they conflict. The analogy illustrates compatibility, but PATR-II structures can be nested and shared, and unification is a precise formal operation rather than a human negotiation over which value to prefer.
Worked example
Consider a tiny teaching grammar whose sentence rule combines a subject noun phrase with a verb phrase and requires their number features to match. Its lexicon contains robot with number singular, robots with number plural, moves with number singular, and move with number plural. The example deliberately considers only these noun and verb uses.
There are four possible two-word combinations. “Robot moves” matches singular with singular; “robots move” matches plural with plural. “Robot move” and “robots moves” produce number conflicts. The constraint therefore licenses two of the four combinations under this stipulated fragment.
If the number constraint is removed while the structural pattern remains, all four combinations fit the category sequence. The difference is caused by the feature requirement, not by a change in word order. This is a hand-checked conceptual illustration, not executable PATR-II syntax or a claim that the fragment models complete English sentences, determiners, imperatives, or every use of move.
Illustrative case study
Fictional case: Alder Linguistics maintains a small grammar for instructional examples. A newly added verb fails in every test sentence. The developer initially suspects the parser, but inspection shows that the lexical entry uses the value sg while the rest of the grammar consistently uses singular.
The team standardizes the feature vocabulary and adds examples covering both successful and conflicting agreement combinations. The case illustrates that atomic labels receive the meaning the grammar assigns them; similar-looking abbreviations do not automatically unify. No performance gain or software execution is claimed, and the correction would still need testing in the actual implementation.
Common misconceptions and mistakes
Do not confuse the context-free skeleton with the expressive power of the entire augmented formalism. Feature structures and their constraints contribute substantial information beyond category sequences.
Do not treat unification as a rule that overwrites whichever value was encountered first. In the core compatibility view, incompatible requirements cause failure rather than an arbitrary choice. Implementation-specific default or priority mechanisms need separate explanation.
Do not assume a successful parse establishes the truth or usefulness of a sentence. It establishes compatibility with the encoded grammar. Coverage gaps, unintended analyses, and inappropriate feature choices remain the responsibility of grammar design and evaluation.
Business use
For most business readers, PATR-II is useful background for understanding explicit grammar-based language tools and legacy linguistic resources. Its practical lesson is that a language analysis can carry inspectable constraints rather than only an unexplained score.
When evaluating a system built with such a grammar, ask which language fragment it covers, how ambiguity is handled, and how lexical or rule changes are tested. Maintenance effort is a real part of the comparison. A transparent constraint is valuable when it captures the intended distinction, but a detailed grammar is not automatically complete or well matched to unrestricted customer language.
People also ask
Is PATR-II a particular parser algorithm?
No. It is a linguistic representation formalism; implementations provide parsing procedures.
Does it require one specific linguistic theory?
Its design aims to support encoding different linguistic analyses rather than prescribe a single theory.
Can incompatible feature values simply be merged?
Not under ordinary unification at the same required feature location. A conflict prevents that combination from succeeding.
Check your understanding
A rule requires subject and verb number to agree, but their specified values are singular and plural. What happens in the toy analysis?
Answer: Unification fails for that combination, so the candidate analysis is not licensed by the rule's feature constraints.
Related terms
- Unification
- Phrase structure grammar
- Lexicon
- Parsing
- feature structure
- declarative grammar
- agreement
Learn next
Work through feature-structure unification with compatible, unspecified, and conflicting values. Then compare the core PATR-II description with an implementation manual so that added syntax and extra operations remain clearly distinguished.
More in Natural language and speech
- Part of speech - A part of speech is a grammatical category used to describe a word's characteristic behavior, such as noun, verb, adjective, or adverb.
- Part-of-speech tagging - Part-of-speech tagging is the task of assigning grammatical category labels to words in context, using a defined tagset.
- Particle - A particle is a grammatical term for certain small function words whose interpretation depends on association with another word, phrase, or construction.
- Period disambiguation - Period disambiguation is the task of deciding what a period character does in its particular context, especially whether it marks the end of a sentence.
- Permutation language modeling - Permutation language modeling trains an autoregressive model using different orders for predicting token positions.
- Personal pronoun - A personal pronoun is a pronoun associated with grammatical person, such as English “I,” “you,” “she,” “he,” “it,” “we,” or “they.
Sources
Primary sources: Shieber, The Design of a Computer Language for Linguistic Information, section 2.1: https://aclanthology.org/P84-1075.pdf ; SIL, PC-PATR Reference Manual, introduction and formalism overview: https://software.sil.org/am/downloads/r/pc-patr/pcpatr.html . Relevant passages were opened and read. This is a historical formalism entry, not a claim about current software availability. The toy grammar and Alder Linguistics case are original illustrations.
Previous: Path planning Next: Pattern grammar