Head-driven Phrase Structure Grammar

The Head -driven Phrase Structure Grammar ( HPSG ) is a grammatical theory that emerged in the 1980s based on the revival of the context-free phrase structure grammars as generative grammar theory of the family of Unifikationsgrammatiken. In HPSG grammatical rules are formulated as constraints that must fulfill correct sentences and phrases; Transformation rules therefore do not apply. The entire information about a linguistic sign is grouped together in a single feature description. Unlike some other theories of grammar word -specific information is given in full in the lexicon, so that only a few grammar rules are necessary.

Like all phrase structure grammars is the HPSG a Konstituentengrammatik. So It is based on the principle of constituency and is therefore not a dependency grammar, which is based on the principle of dependence.

  • 5.1 general surveys and introductions
  • 5.2 Specific issues in
  • 6.1 Overview bibliographies, other materials
  • 6.2 implementations


The Head - Driven Phrase Structure Grammar was developed by Carl Pollard and Ivan Sag from the mid-1980s. Major components were older syntax theories, particularly non- derivational approaches, encouraged or accepted, for example, the categorial (CG ), Generalized Phrase Structure Grammar (GPSG ), Arc Pair Grammar ( AC), Lexical - Functional Grammar (LFG ), but also the prevailing Government and Binding theory ( Government and Binding theory, UK) Noam Chomsky. The presentation of the semantics is based on parts on the situation semantics; formal foundations of computer science come from. The first comprehensive presentation of the theory gave Carl Pollard and Ivan Sag with the published 1987 Book Information -Based Syntax and Semantics, Volume I ( Pollard, Sag 1987); a revised version they made in 1994 before the factory Head - Driven Phrase Structure Grammar ( Pollard, Sag 1994). Right from the beginning, other scientists Head - Driven Phrase Structure Grammar have taken and proposed modifications, extensions and applications in different languages. On many issues, there is thus a variety of views expressed by different scientists to describe different languages. Some theories of grammar that occupy a middle position between HPSG and other theories have been developed, such as the Sign -Based Construction Grammar, which takes up ideas of Construction Grammar within the HPSG formalism.


Basic Concepts

In HPSG, all words and phrases can be modeled as a sign within the meaning of Ferdinand de Saussure, that is, as the form - meaning pairs. Syntactic properties, the sound structure and the meaning of a sign can be presented in a single attribute-value matrix, hence HPSG is considered monostratal. The attribute-value matrix of each character includes at least one feature PHON, representing the phoneme, and a value SYNSEM, which summarizes in a matrix of type synsem information about grammatical properties and the meaning. There are also suggestions for the formal representation of other aspects of a language in attribute-value matrices, such as word order (see the section word order ) and the syllable structure.

Unlike many other theories of grammar HPSG is declarative: the entire grammar including the lexicon is formulated as a description grammatically correct characters. Therefore, in HPSG, there are no rules for change or movement of constituents. Instead, grammatical rules exclusively in the form of constraints expressed that must be met by well-formed characters. An example is the congruence of a verb with its subject. While in non- restriction -based theories of grammar such as the transmission of a feature from the subject to the verb is assumed to possess in HPSG verb and subject corresponding features, which must be equal to certain restrictions in both characters.

HPSG is also a lexicalized grammar theory to a large extent, that is, the grammatical information is saved to a large extent in the lexicon, the grammar itself needs to provide only a few restrictions for the processing of the lexicon. For example, the arguments of a verb in lists are defined, which are included in the characterization of the verb; the grammar then determines by restrictions, such as the arguments are realized.

Formal Foundations

All information about a characters are specified in HPSG in a hierarchically structured attribute-value matrix ( attribute-value matrix, short AVM). In each line, thereby, the corresponding value is given for a particular attribute. Each value has a type and it can have its own characteristics. The type determines what are the characteristics of an object and which types have the appropriate values. For example, in the formalism of Pollard and Sag 1994 each object of type LOCAL synsem a feature with an object of type localhost as a value and a feature NONLOC with a value of type nonloc. The types form a hierarchy, with subtypes inherit the features of their supertypes. Types are usually displayed in italics in the graph on the left. The object represented by the following matrix, for example, has the index type and the characteristics PERSON, NUMBER and GENDER. The associated values ​​of the types 2, sg fem and here and have no own characteristics:

Also lists and sets of objects are allowed as values. So the feature SUBCAT required as value a list of objects that have synsem the type:

In the graphical representation of attribute-value matrices is taken into account that usually only necessary for a particular issue features a matrix is presented. In addition, longer paths in the literature are often | abbreviated to " ". Therefore, the following two matrices are equivalent:

An HPSG - based description of a language has at least the following formal elements:

  • A signature, in which is fixed, what types are available, what features they have and what types have their values
  • Principles, formulate the constraints which must apply to all well-formed characters
  • Functions and relations, for example, to calculate morphological forms and for joining lists

The lexicon is expressed either by restrictions on the type or word but it is given a separate status outside the restrictions.

Restrictions can be formulated with the help of unterspezifierten feature descriptions, which must be in the application of the limitation to a feature description with this unifiable then so that the constraint is satisfied. As an example, serve the head feature principle, which states that in any phrasal sign that has a head, the HEAD value must be equal to that of the head daughter. This can be formulated as an implication ( is the use of quantifiers in the HPSG literature is not mandatory):

The lexicon

The lexicon consists of HPSG descriptions for the words of a language, the so-called lexical entries. For this purpose, a disjunction of feature descriptions for each word are used:

In order to allow generalizations, words can be divided into word classes, which cover those features which are all words one word class together. To fulfill nouns such as woman, sun, cat's general scheme for feminine nouns in the singular: in the analysis of Müller in 1999, for example, their head features are noun type, they require an item as a complement, and have the same values ​​for person, number and gender.

Some syntactic and morphological phenomena are mainly captured by so-called lexical rules, license the lexicon entries by putting them with other entries in relationship. For example, passive verb forms could be licensed in a simplified grammar by a passive verb is licensed to each transitive verb whose subject matches the object of the transitive verb. In many HPSG -based theories while the operator " ↦ " is applied, which combines the description of the output word and licensed by the control word:

Are lexical rules - depending on the theoretical approach - either as meta-rules for the description of the lexicon or as restrictions on words within the formalism of HPSG formulated.

Semantics and Pragmatics

The meaning of a sign is specified in a synsem object in a matrix, which is usually called CONTENT and usually has one of several subtypes of content each with its own characteristics. Pollard and Sag 1994, for example, see the types Psoas, nom - obj and quant ago, Bender, Sag and Wasow 2003, around the other hand, take a uniform feature geometry for all CONTENT ( with them SEM ) values.

To illustrate the situation semantics is in most HPSG theories on so-called parameterized issues (English parametrized state of affairs, short Psoas ) as possible, which are represented by matrices of type Psoas. Parameterized facts consist of a relation as seen beat, specify book, Man and parameters, the different semantic roles in the relation. In HPSG the parameterized issues are also shown in attribute-value matrices. This allows the relation " the man sees the dog " according to Müller represented by the following matrices 2007 ::

The first matrix is the Psoas representation of the relation seen with two arguments. The second matrix represents the CONTENT value of the description of man dar. your index is identified by the day with the agent of see the RESTRICTIONS amount determines that it is in a man. Analog is the case with the third matrix, which is located in the description of dog.

By dividing structure, the semantic roles can also connect with syntactic functions. The following excerpt from the LOCAL value of "see" identifies the indices of the two noun phrases in the SUBCAT list (see the section complements) with the agent or patientive ( feature geometry according to Müller, 2007):

In recent HPSG -based theories are also other theories such as minimum Recursion Semantics (MRS ) and Lexical Resource Semantics (LRS ) application that can be represented by attribute-value matrices as well.

Context information in a matrix of type context in the path SYNSEM | CONTEXT indicated that characteristics such as BACKGROUND and C- INDICES has | LOC. BACKGROUND is a lot of Psoas objects that provide background information about the utterance of the sentence. C- INDICES has several attributes that give in the form of indexes information about the circumstances of speaking, such as the speaker, the addressee and the place.

Structures with heads

In HPSG is usually assumed that phrasal signs are composed of a head daughter and a certain number of non- head daughters. The composition of such a structure is mainly determined by grammar principles and characteristics of the daughters. Especially the early research tried to make do with the smallest possible number of very general types of structures, so-called ID - schemes ( Immediate Dominance schemas ). In the grammar by Pollard and tip 1994, there are six ID schemes, including, for example, one for the connection of the head and of a complement, and for the connection of head and adjunct. Younger versions, especially those who are close to the construction grammar, often contain very numerous and specific designs. So Ginzburg and Sag develop 2000 23 ID schemes. Opinions were divided as to whether certain phenomena with ID schemes should be described, which do not have non- head daughter, but only one head daughter. Such proposals have been made, for example, the introduction of non-local information without the use of tracks ( see Non-local information ). The following section focuses more on approaches that Pollard and Sag 1994, following start from a small number of ID - schemes and a strong lexicalization.

Composition of the semantics

In HPSG versions following the Pollard and Sag 1994, it is assumed that the semantics of a phrase in most cases is the same as the head daughter. However, the semantics is to be determined by the Adjunkttochter if it is a structure with adjunct. For the head - adjunct structure " red book " and her daughters thus SYNSEM following values ​​result:

  • " Red book "
  • " Book":
  • " Red":

The feature HEAD contains information that is common to the head and its phrasal projections, such as case, number and gender in noun phrases. The head feature principle conditionally with this is that the HEAD feature of a phrase with its head is the same.

Types of non- head daughters

Depending on the approach, different types of non- head daughters are distinguished. This paragraph can only provide examples.


Komplementtöchter are determined by their lexical head in the valence. Information on the valence of a character to be in one or more lists, such as the SUBCAT characteristic in the path SYNSEM | stored LOC. Usually they contain SYNSEM objects of arguments of the character that has not yet set. Depending on the theoretical approach could be for a verb like so "see" the following extracts from a characterization to formulate:

  • With SUBCAT feature
  • With separate lists for subject and complements

A grammatical principle defines it as complements are ligated. Assuming a SUBCAT list, it can be formulated as follows:


Following Pollard and Sag 1994 selegieren cofactor their heads. To this end, they get a head feature MODIFIED, short MOD, which is identified by structure sharing with the SYNSEM value of the head as an ID system defines are defined in the head - adjunct structures. In addition, when the situation semantics is used (see semantics and pragmatic ) assumed that the semantics of the Adjunkttochter is identical to the parent, so this restriction is included in the semantic principle. This makes it possible to capture the semantics of phrases with a so-called capsule Direction modification. A number of recent works, however, is the assumption that adjuncts as complements are determined by a separate Valenzliste of the head; This approach is also known under the name of adjunct -as- Complement Approach.


For the analysis of various phenomena numerous other species were introduced by non- head daughters. So Pollard and Sag 1994 did have its own word class proposed for words as the English conjunction, the so-called marker (English marker). According to its head - marker schema selegieren marker on the SPEC feature SYNSEM the value of the head, in addition to them and with their mother, the feature MARKING has marked the value.

A special feature is the formal introduced in 1994 by Pollard and Sag analysis of the quantifiers out its Skopuseigenschaften be modeled with a Cooper - Store, which contains the semantics of the quantifier and is passed upwards from the quantifier out until he tied due to a limitation will.

Non-local information

For connections between nodes that are farther away from each other in the feature description of the parent phrase HPSG uses so-called non-local features (English: unbounded dependency ) for the remote connection will contain necessary information and passed from node to node to provide information in the two relevant to provide nodes. A grammatical principle is ensuring that non-local values ​​are passed on until they are tied for a reason.

This allows, for example, the analysis of the extraction of noun phrases. In many HPSG approaches as is assumed in other theories of grammar that the corresponding noun phrase to its actual position one track ( english trace) leaves, which is koindiziert with the extracted phrase and of other words differs in that it PHON value is empty. Examples from English are:

  • John1, Mary loves _1
  • I wonder who1 Sandy loves _1

Pollard and Sag 1994 and others have proposed a feature description for traces after their LOCAL value in which all the local information stored on syntax and semantics, is identified with an element of the nonlocal SLASH list, which then so by the mentioned grammar principle is passed on until it is bonded again. Other analyzes use phrases with only one daughter or lexical rules to introduce the non-local information to a node without requiring a blank sign.

Attachment Theory

Binding theories make statements as to whether noun phrases are realized as reflexive as personal pronouns or as Nichtpronomina. The proposed bond theories are assuming that two conditions must be met before a noun phrase anaphoric, in the classic grammar so as reflexive, appears:

  • The noun phrase must be consistent with another in a certain relationship
  • These two noun phrases must have the same value INDEX

The problem of defining this relationship is achieved by the various approaches differ significantly in HPSG is always that the anaphoric noun phrase, her mother, or their projection is oblique than the other noun phrase, with which it is koindiziert.

Constituent order

The constituent order of a language can be expressed in HPSG other restrictions, the so-called Linear Precedence rules, short LP- rules, which form the

Have. X Y is the reverse position. In languages ​​in which heads are at the end of a phrase, so the rule is

Languages ​​with more complex or free word order as the German need complicated rules, for example in the form of a disjunction of several LP rules.

For languages ​​with free word order more particularly analyzing proposals that go beyond the formulation of complex LP- rules exist. A free arrangement of the complements was declared by some scholars with lexical rules for rearranging the Valenzlisten or by adopting disordered Valenzmengen instead of lists, flat structure trees were considered. In a series of works is proposed, however, does not remove elements of Valenzlisten after setting, but continue to reach upward to prevent characters from another location can still access it.

Another approach assumes that the word order is not directly related to the syntactic structure, ie that constituents need not be continuous. For this purpose, it is assumed that the daughters of a phrasal sign are collected in a so-called linearization domain in the feature DOM, then LP- rules are applied to it. In the union of two domains, they can be joined to the domain of the mother with the shuffle operation in which the position of the characters from a domain maintained relative to each other; a domain can also be compact, so can be no strange characters between characters that domain.

Implementation and application in computational linguistics

Since the beginning of the 90s of the 20th century, various systems for implementing HPSG grammars have been developed in computational linguistics. Only a minor part of the HPSG - based architectures is directly to the formalism by applying the requirements formulated in the theory of constraints to each linguistic object. As might be subject to efficiency problems in such systems, other implementations encode a portion of the restrictions as relations with linguistic objects as arguments, with particular phrase structure rules play an important role, which is not present, although in HPSG, but facilitate efficient parsing. Some implementations also allow parsing with discontinuous constituents, which play a role in certain HPSG grammars (see word order ). HPSG -based systems play a role in research in the field of Deep Processing, where they can also be combined with methods of Shallow Processing.