Part I
Foundations
Everything in this index is a statement about labelled trees. This part defines those trees from first principles: graphs, then rooted ordered trees, then the Function:Category labelling of CGEL, then the HPSG constraints that decide which trees are well formed.
Concept entries cite these items under Builds on only where a concept genuinely presupposes one. Apart from that, the concepts do not form a dependency order: the general rules of the grammar are parallel, each an independent instance of one of the schemata (Definition 21), so they are simply listed in the index.
- Graph
- Directed graph (digraph), and how we use it
- Rooted tree
- Ordered tree
- Dominance
- Precedence
- Constituents, branching and unary chains
- The labelled constituent tree (Function:Category)
- Bracketed tree notation
- Lexical categories (word classes)
- Phrasal categories
- Grammatical functions
- Heads and headedness
- Fusion of functions
- Gaps, fillers and coindexation
- Coordination is not headed
- The HPSG type hierarchy
- Features and feature structures
- Head Feature Principle
- Valence: SPR, SUBJ and COMPS
- Schemata
- Licensing and well-formedness
Definition 1 Graph
A graph is a pair $G = (V, E)$ where $V$ is a finite, non-empty set of vertices (we say nodes) and $E$ is a set of edges, each edge being a two-element subset $\{u, v\} \subseteq V$.
Two nodes $u, v$ are adjacent when $\{u,v\} \in E$. A path from $u$ to $w$ is a sequence of distinct nodes $u = v_0, v_1, \dots, v_k = w$ with each $\{v_{i}, v_{i+1}\} \in E$. A graph is connected when every two nodes are joined by a path. A cycle is a sequence of at least three distinct nodes, with edges joining successive nodes and joining the last back to the first; a graph is acyclic when it has no cycle.
Graphs by themselves say only which things are linked. Grammar needs more: which element contains which, and in what order the words come. The next definitions add exactly those two pieces of structure.
V = {a, b, c}, E = { {a,b}, {a,c} } a — b
|
cPermanent link https://grammar.canonical-english.org/f/d70780b4-4daf-49ac-bca8-9e3232fb829e
Definition 2 Directed graph (digraph), and how we use it
A directed graph or digraph is a pair $D = (V, A)$ where $A \subseteq V \times V$ is a set of ordered pairs called arcs. An arc $(u, v)$ goes from $u$ to $v$; we write $u \to v$.
How we use digraphs. In every syntactic structure on this site there is one arc $u \to v$ exactly when $v$ is an immediate constituent of $u$: the arc points from a whole to one of its parts (mother to daughter). Nothing else is an arc. In particular:
- Linear order of words is not encoded by arcs; it is a separate relation (precedence, below).
- The link between a gap and its filler is not an arc; it is a separate relation, coindexation (see Gaps), so the constituent digraph stays a tree.
- Function and category are labels on nodes, not on arcs. (Equivalently, the function of $v$ could be read as the label of the unique arc into $v$; we keep it on the node.)
The in-degree of $v$ is $|\{u : u \to v\}|$ and its out-degree is $|\{w : v \to w\}|$.
"Kim left": Clause → NP, Clause → VP, NP → Nom, Nom → N, VP → V
Permanent link https://grammar.canonical-english.org/f/1a6bc8c7-033d-4a9b-82d5-02be850eadca
Definition 3 Rooted tree
A rooted tree is a digraph $T = (V, A)$ with a distinguished node $r \in V$, the root, such that
- $r$ has in-degree $0$;
- every other node has in-degree exactly $1$;
- every node is reachable from $r$ by a directed path.
For an arc $u \to v$ we call $u$ the mother (parent) of $v$ and $v$ a daughter (child) of $u$. Daughters of the same mother are sisters. A node with out-degree $0$ is a leaf; any other node is internal. These conditions imply that $T$ has no cycles and that the path from $r$ to any node is unique.
In a syntactic tree the root is the whole sentence (or whatever expression is being analysed) and the leaves are its words and its gaps.
Permanent link https://grammar.canonical-english.org/f/f93be5f4-81a2-487b-9f57-61be51724a9b
Definition 4 Ordered tree
An ordered tree is a rooted tree together with, for every internal node $u$, a strict total order $<_u$ on the daughters of $u$ (the left-to-right order in which they are spoken or written).
The daughter orders extend to a single order on the leaves: list the leaves by a depth-first, left-to-right walk from the root. In a syntactic tree that order is the order of the words in the sentence.
English syntax needs order: Kim saw Pat and Pat saw Kim have the same unordered tree shape but different ordered trees, and different meanings.
Permanent link https://grammar.canonical-english.org/f/a3ff869a-d11d-48a4-a37c-97165ee15756
Definition 5 Dominance
In a rooted tree, node $u$ dominates node $v$, written $u \trianglerighteq v$, when there is a directed path (possibly of length $0$) from $u$ to $v$. Dominance is reflexive, antisymmetric and transitive: a partial order on $V$. The root dominates every node.
$u$ properly dominates $v$ ($u \triangleright v$) when $u \trianglerighteq v$ and $u \neq v$; $u$ immediately dominates $v$ when $u \to v$ is an arc.
Grammatical statements are mostly about immediate dominance (what a phrase directly consists of) and occasionally about longer paths (for example, how far a gap may be from its filler).
Permanent link https://grammar.canonical-english.org/f/fb58b014-74a0-4b46-a5f9-0aee4c2816b9
Definition 6 Precedence
In an ordered tree, node $u$ precedes node $v$, written $u \prec v$, when neither dominates the other and every leaf dominated by $u$ comes before every leaf dominated by $v$ in the leaf order.
Exclusivity. For any two distinct nodes exactly one of these holds: $u \triangleright v$, $v \triangleright u$, $u \prec v$, $v \prec u$. So two constituents either nest or do not overlap at all; constituents never partly overlap. This is what makes a tree a bracketing of the sentence.
Permanent link https://grammar.canonical-english.org/f/cf54a7f8-cad9-4a5c-beac-a12da95d7f46
Definition 7 Constituents, branching and unary chains
The constituent of node $u$ is the subtree of all nodes $u$ dominates; its yield is the sequence of its word leaves. An expression is a constituent of a sentence when it is the yield of some node.
A node is branching when it has two or more daughters and unary when it has exactly one. A maximal path of nodes linked by unary arcs, $u_1 \to u_2 \to \dots \to u_k$, is a chain; its first node is the chain's top. Chains record that one expression is several things at once: in Kim left, the word Kim is a noun ($N$) that alone forms a nominal ($Nom$) that alone forms a noun phrase ($NP$).
NP ─ Nom ─ N "Kim" (one chain of three nodes; the NP is its top)
Permanent link https://grammar.canonical-english.org/f/5b2b9213-d58b-4e77-b2ef-edd5d826eeaa
Definition 8 The labelled constituent tree (Function:Category)
A syntactic tree is an ordered tree $T$ with two labelling maps
where $\mathcal{C}$ is the set of categories and $\mathcal{F}$ the set of functions (both enumerated below). Every node has a category; every node except the root has a function. We write a node's labels as $\textit{Function}{:}\textit{Category}$, for example $\mathrm{Subj{:}NP}$ (a noun phrase functioning as subject), $\mathrm{Head{:}N}$, or $\mathrm{Head{:}V}$ (the verb that heads a verb phrase, which CGEL calls the Predicator).
The two labels answer different questions. The category says what kind of expression a node is, independently of where it occurs. The function says what job it does in its mother. The same category can do many jobs (an NP can be subject, object, predicative complement, ...), and the same job can be done by different categories (a subject can be an NP or a clause). Keeping the two apart is the central design principle of the CGEL framework.
A word leaf additionally carries its form (and in Canonical English its lexicon sense). Additional conditions follow in Fusion and Gaps.
Permanent link https://grammar.canonical-english.org/f/8624f13b-f1d5-41a4-a64e-1f4f561a4f83
Notation 9 Bracketed tree notation
We write trees linearly. A node is written
with daughters in order; the root omits the function. A word leaf is Function:Category"word". A filler carries {index=x} after its label (after the word, for a leaf); a gap is the leaf Function:GAP{index=x}. The ASCII drawings on concept pages show the same tree vertically.
Formally the notation is the language of the grammar
node ::= [Function ":"] Category [ '"' word '"' ] [ "{index=" id "}" ] [ "[" node ("," node)* "]" ]
and every string it generates denotes exactly one labelled ordered tree. This is the notation of noah/FORMAT.md and of CGELBank, the treebank in the CGEL framework.
"Kim saw the dog."
Clause[ Subj:NP[ Head:Nom[ Head:N"Kim" ] ],
Head:VP[ Head:V"saw",
Obj:NP[ Det:DP[ Head:D"the" ], Head:Nom[ Head:N"dog" ] ] ] ]
Clause
├── Subj:NP ── Head:Nom ── Head:N Kim
└── Head:VP
├── Head:V saw
└── Obj:NP
├── Det:DP ── Head:D the
└── Head:Nom ── Head:N dogPermanent link https://grammar.canonical-english.org/f/1e9a6dd8-8987-4c9d-85ac-036060178d1f
Definition 10 Lexical categories (word classes)
The lexical categories $\mathcal{C}_{\mathrm{lex}} \subset \mathcal{C}$ are the categories of word leaves. A word is assigned a category by its grammatical behaviour (inflection, what it combines with, where it can stand), not by meaning alone. Following CGEL, determinatives (the, some, this) are a word class of their own, distinct from adjectives; pronouns are a subclass of nouns; and prepositions include words traditionally called adverbs or conjunctions when they head phrases in the same way (before in before lunch and before you go).
| Label | Category |
|---|---|
N | noun |
N_pro | pronoun |
V | verb |
V_aux | auxiliary verb |
Adj | adjective |
Adv | adverb |
P | preposition |
D | determinative |
Sdr | subordinator |
Coordinator | coordinator |
Int | interjection |
Convention (leaves). In the notation used here, lexical nodes carry words, not daughters. The full treebank also has special structures for multiword names and compounds; the worked trees here use the simpler word-leaf convention.
Permanent link https://grammar.canonical-english.org/f/6f36c2b2-33ee-4f8c-bf3d-dfeb3b926d80
Definition 11 Phrasal categories
The phrasal categories $\mathcal{C}_{\mathrm{phr}}$ are the categories of internal nodes. Most are projections of a word class: a phrase whose head is (ultimately) a word of that class. CGEL adds an intermediate level for nouns, the nominal ($Nom$): the noun with its internal modifiers and complements but without the determiner, so that in the old dog the string old dog is a constituent.
| Label | Category |
|---|---|
Clause | clause |
Clause_rel | relative clause |
NP | noun phrase |
Nom | nominal (a noun with its internal dependents, below the determiner) |
VP | verb phrase |
AdjP | adjective phrase |
AdvP | adverb phrase |
PP | preposition phrase |
PP_strand | preposition phrase whose object is a gap at its end |
DP | determinative phrase |
IntP | interjection phrase |
Coordination | two or more coordinated constituents |
NP+PP | coordinate of an NP and a PP |
NP+AdvP | coordinate of an NP and an AdvP |
NP+AdjP | coordinate of an NP and an AdjP |
NP+Clause | coordinate of an NP and a clause |
$\mathrm{GAP}$ is a special category for empty leaves (see Gaps). The category inventory above includes treebank notation not yet executable in our grammar, notably PP_strand and the mixed-category coordinations. Conversely Word in grammar.toml is a compiler wildcard for a cited expression, not an additional linguistic word class. $\mathcal{C} = \mathcal{C}_{\mathrm{lex}} \cup \mathcal{C}_{\mathrm{phr}} \cup \{\mathrm{GAP}\}$, and the three sets are disjoint.
Permanent link https://grammar.canonical-english.org/f/bd10d638-dab4-4b7b-bbb1-a41cd7bd3ad2
Definition 12 Grammatical functions
The functions $\mathcal{F}$ name the relation between a node and its mother. They are relational: a node is a subject of a clause, a determiner of a noun phrase. The full inventory we use (the labels of CGELBank, written with their CGEL names):
| Label | Function | What it is |
|---|---|---|
Head | Head | the daughter that determines the category of its mother; in a VP the verb, called the Predicator |
Subj | Subject | subject of a clause |
Obj | Object | object |
Obj_dir | Direct object | direct object when a verb has two objects |
Obj_ind | Indirect object | indirect object when a verb has two objects |
PredComp | Predicative complement | complement that describes or identifies a predicand |
Comp | Complement | any other complement licensed by the head |
Comp_ind | Indirect complement | complement licensed by a word other than the head |
Particle | Particle | particle complement, as out in take it out |
Mod | Modifier | modifier, including adjuncts in the clause |
Det | Determiner | determiner of a noun phrase |
Det-Head | Determiner-head (fused) | one constituent that is both determiner and head |
Mod-Head | Modifier-head (fused) | one constituent that is both modifier and head |
Marker-Head | Marker-head (fused) | one constituent that is both marker and head |
Head-Prenucleus | Head-prenucleus (fused) | one constituent that is both head and prenucleus |
Marker | Marker | word that marks a construction, such as that before a clause |
Coordinate | Coordinate | one constituent of a coordination |
Supplement | Supplement | loosely attached material set off by intonation or punctuation |
Prenucleus | Prenucleus | constituent before the nucleus of its clause, such as a fronted wh-phrase |
Postnucleus | Postnucleus | constituent after the nucleus of its clause |
DisplacedSubj | Displaced subject | subject placed after the verb |
ExtraposedSubj | Extraposed subject | subject at the end of the clause with it in subject position |
Vocative | Vocative | name or description of the person addressed |
Flat | Flat | part of a name or number with no internal structure |
Compounding | Compounding | constituent inside a compound word |
Obj+Mod | Object+modifier | coordinate whose parts have different functions |
Complements versus modifiers. A complement (Subj, Obj, PredComp, Comp, ...) must be licensed by its head: devour requires an object, sleep does not take one. A modifier (Mod) is not selected by the head and can generally be added or dropped freely. This distinction runs through every phrase type.
Permanent link https://grammar.canonical-english.org/f/b3f4ec5c-1c48-44cd-9837-7a94f70e5486
Principle 13 Heads and headedness
Axiom (one head). Every internal node other than a coordination (including a mixed-category coordination such as NP+PP) has at most one daughter whose function is $\mathrm{Head}$ or a fused function containing Head (see Fusion). A Clause whose daughters are a Prenucleus and a Clause is still headed by that inner Clause.
The head is the daughter that determines the category and the essential grammatical properties of its mother: a VP is headed by a verb, an NP by a nominal, a nominal by a noun. Following the chain of heads downward from any phrase reaches a single word, the ultimate head (CGEL's term), e.g. dog in the old dog that barked.
Endocentricity. If $u$ has head daughter $h$, then $\mathrm{cat}(u)$ is a projection of $\mathrm{cat}(h)$ (for example $NP \leftarrow Nom \leftarrow N$, $VP \leftarrow V$, $Clause \leftarrow VP$). The HPSG layer turns this into the Head Feature Principle.
NP[ Det:DP[ Head:D"the" ], Head:Nom[ Mod:AdjP[ Head:Adj"old" ], Head:Nom[ Head:N"dog" ] ] ] heads: NP → Nom → Nom → N "dog" (ultimate head: dog)
Permanent link https://grammar.canonical-english.org/f/c396b226-a74d-487e-b2c9-242e1f82352a
Convention 14 Fusion of functions
Sometimes one constituent does two jobs at once. In Many left, the word many is both the determiner of the noun phrase and its head: there is no separate noun. CGEL calls this fusion, and we represent it with a single node carrying a fused function
Convention. A fused function $X\text{-}Y$ assigns both functions to one node; for the one-head axiom it counts as the head of its mother. We do not postulate an empty head beside a determiner. Fusion is different from ellipsis, where a head is missing but recoverable from context.
"Many left." Clause[ Subj:NP[ Head:Nom[ Det-Head:DP[ Head:D"many" ] ] ], Head:VP[ Head:V"left" ] ] "the rich" NP[ Det:DP[ Head:D"the" ], Head:Nom[ Mod-Head:AdjP[ Head:Adj"rich" ] ] ]
Permanent link https://grammar.canonical-english.org/f/bc68b307-6fe2-4c22-9123-d9cfffb4a6c7
Definition 15 Gaps, fillers and coindexation
In the book that you read \_, the object of read is missing from its usual position and is understood as book. We model this with a gap: a leaf $g$ with $\mathrm{cat}(g) = \mathrm{GAP}$, a function (here Obj) and an index. The constituent it is understood as, the filler, carries the same index.
Axioms (coindexation). Let $I$ be the set of indices in a tree.
- Every gap carries exactly one index; no node carries more than one.
- Each index carried by a gap is carried by exactly one non-gap node, its filler, which is the top of its chain.
- Every index on a non-gap node is carried by at least one gap.
Coindexation is a relation alongside the tree, not an arc of it, so the structure remains a tree. The path from a filler down to its gap can be arbitrarily long (the book that I think you said you read \_): this is what makes such dependencies unbounded.
"the day that he left"
NP[ Det:DP[ Head:D"the" ],
Head:Nom[ Head:N"day"{index=x},
Mod:Clause_rel[ Marker:Sdr"that",
Head:Clause_rel[ Subj:NP[ Head:Nom[ Head:N_pro"he" ] ],
Head:VP[ Head:V"left", Mod:GAP{index=x} ] ] ] ] ]Permanent link https://grammar.canonical-english.org/f/962c5b50-2f0c-4493-bd2d-a03fff3d606a
Convention 16 Coordination is not headed
A node of category $\mathrm{Coordination}$ has two or more daughters with function $\mathrm{Coordinate}$ and no head. A coordinator (and, or, but) is a $\mathrm{Marker{:}Coordinator}$ inside the coordinate it introduces. The coordinates are of equal status; none of them determines the category of the whole by itself. Coordination of unlike categories gets a combined label such as $NP{+}PP$ in the treebank inventory. The executable HPSG grammar currently supports fewer combinations than the notation can express.
"Kim and Pat"
Coordination[ Coordinate:NP[ Head:Nom[ Head:N"Kim" ] ],
Coordinate:NP[ Marker:Coordinator"and", Head:NP[ Head:Nom[ Head:N"Pat" ] ] ] ]Permanent link https://grammar.canonical-english.org/f/0e0f9af3-1a09-4832-8f55-d06e47c4dfd4
Definition 17 The HPSG type hierarchy
The trees above say what the structure is. The HPSG layer (Head-driven Phrase Structure Grammar) says which structures are allowed, by attaching information to nodes and stating general constraints. Its first ingredient is a type hierarchy: a finite partial order $(\mathcal{T}, \sqsubseteq)$ where $\sigma \sqsubseteq \tau$ reads "$\sigma$ is a subtype of $\tau$". Each phrasal and lexical category is mapped to a head type.
In noah/src/noah/grammar.toml the head types are
A type inherits every constraint of its supertypes: everything said of agreeing (it carries the feature AGR) holds of nouns, determinatives and verbs.
Permanent link https://grammar.canonical-english.org/f/b24a0e96-42a6-445c-9629-a4e1ec486046
Definition 18 Features and feature structures
A feature structure assigns values to features (attributes). In our compact layer each node $v$ carries a finite partial function $\varphi_v : \mathbf{Feat} \rightharpoonup \mathbf{Val}$, e.g.
We write features in brackets after a category, as in $\mathrm{NP}[\mathrm{CASE}{=}\mathit{acc}]$ or $\mathrm{V}[\mathrm{VFORM}{=}\mathit{past}]$. Two structures unify when they agree on every feature both define; the result is their union. Grammaticality checks reduce to unification: this dogs fails because the determinative has $\mathrm{AGR}{=}\mathit{3sg}$ and the nominal $\mathrm{AGR}{=}\mathit{3pl}$.
The features of noah/src/noah/grammar.toml and their values:
| Feature | Values |
|---|---|
AGR | 1sg, 2sg, 3sg, 1pl, 2pl, 3pl |
CASE | nom, acc |
OBJ | trans, intrans, clause, speech |
VFORM | plain, pres, past, pastpart |
AUX | be, modal, do, have |
ADV | degree, manner, frequency, negation, place, time, focus, linking |
SBR | decl, interrog, infin |
COORD | and, or, but, nor |
DTYPE | cardinal, noncardinal, personal |
COUNT | count, noncount |
DEF | def, indef |
DFUNC | det, predet, fused |
PFORM | of, nonof |
GEN | gen |
PROPER | proper |
CLAUSE | main, bare, content |
NOMTYPE | pronoun, fused, word, phrase, counted, proper |
VPTYPE | word, phrase, auxneg |
LIKE | NP, VP, Clause |
MARK | quote, paren |
The full textbook form of HPSG nests these under SYNSEM and LOCAL and shares values with tags; ours keeps them flat.
Permanent link https://grammar.canonical-english.org/f/d002410c-a1ea-4afe-900b-737c6e536f5a
Principle 19 Head Feature Principle
Principle. If node $u$ has head daughter $h$, then for every head feature $f$ of $\mathrm{type}(\mathrm{cat}(u))$,
So the properties that make a phrase the kind of phrase it is are inherited from its head: a VP is tensed because its verb is, an NP is plural because its noun is. Together with Headedness this links the tree's head chains to its features.
Permanent link https://grammar.canonical-english.org/f/ad112a93-3b41-4bc4-aabc-78c5791d230e
Principle 20 Valence: SPR, SUBJ and COMPS
A word lists what it must combine with. COMPS lists its complements (from its lexical frame: OBJ = trans gives $\langle NP[\mathrm{CASE}\ acc]\rangle$, OBJ = intrans the empty list). SPR names a nominal's specifier (the determiner of an NP), which must agree with it in AGR and COUNT. SUBJ names a VP's subject, a nominative NP that agrees with it in AGR.
Valence Principle. A phrase built by combining a head with an item from one of its lists has that item removed from the list; all other list items pass up unchanged. A phrase is saturated when its lists are empty: an NP is a saturated nominal, a Clause a saturated VP.
Permanent link https://grammar.canonical-english.org/f/e056f5ef-a1e5-4e19-a458-22435a361288
Definition 21 Schemata
A schema is a general template for a local tree (a mother and its daughters). Every rule in grammar.toml is an instance of exactly one schema:
| Schema | Local tree | Example |
|---|---|---|
| head-subject | $\mathrm{Clause} \to \mathrm{Subj{:}NP}\ \ \mathrm{Head{:}VP}$ | Kim left |
| head-complement | $\mathrm{VP} \to \mathrm{Head{:}V}\ \ \mathrm{Obj{:}NP}$ / $\mathrm{Comp{:}X}$ | saw the dog |
| head-specifier | $\mathrm{NP} \to \mathrm{Det{:}DP}\ \ \mathrm{Head{:}Nom}$ | the dog |
| head-adjunct | $X \to \mathrm{Mod{:}Y}\ \ \mathrm{Head{:}X}$ (either order) | old dog, left early |
| head-marker | $X \to \mathrm{Marker{:}Sdr}\ \ \mathrm{Head{:}X}$ | that it rained |
| head-only | $X \to \mathrm{Head{:}Y}$ (a unary link in a chain) | $NP \to Nom$ |
| fused-head | $Nom \to \mathrm{Det\text{-}Head{:}DP}$ | many in Many left |
| coordination | $\mathrm{Coordination} \to \mathrm{Coordinate{:}X}^{n},\ n\geq 2$ | Kim and Pat |
Each schema is constrained by the Head Feature Principle and the Valence Principle; a rule adds only what is specific to its construction.
Permanent link https://grammar.canonical-english.org/f/aaef08f7-8549-4966-95d2-51dbbf435d5b
Definition 22 Licensing and well-formedness
A labelled tree $T$ is licensed by a grammar $G$ when
- every word leaf matches a lexicon entry of its category;
- every local tree (an internal node with its daughters, their functions and categories) instantiates a rule of $G$, and hence one of the schemata;
- the feature structures satisfy the Head Feature Principle, the Valence Principle and each rule's own constraints, i.e. all required unifications succeed;
- the root has the start category ($\mathrm{Clause}[\mathrm{CLAUSE}\ \mathit{main}]$) and the coindexation axioms hold.
This defines the intended syntactic acceptance condition for Canonical English, not a claim that the current implementation covers every construction. Local rule matching and feature checks are executable for a subset of these structures; comprehensive gap licensing and broader CGEL coverage remain work in progress. Meaning, discourse, morphology and punctuation entries also need constraints beyond constituent trees. Their prose may explain a concept for which there is not yet an executable HPSG rule.
Permanent link https://grammar.canonical-english.org/f/7ef88fad-5e4b-40f4-97ee-5fc98960334c