Dependency Syntax Notes

Syntax

• Syntax is the study of sentence structure, answering "Who does what to whom?"
• Various theories exist with commonalities: Government and Binding (GB), Minimalist Program (MP), Head-driven phrase structure grammar (HPSG), Lexical Functional Grammar (LFG), Categorial Grammar, and Dependency Grammar.

Why Syntax Matters

• Theoretical syntacticians focus on grammaticality.
• Relevant for NLP applications like text generation and grammar checking.
• Parsing provides scaffolding for semantic analysis, aiding opinion mining, information extraction, and machine translation.

Basic Principles of Syntax

• Form vs. Function
  • Syntactic form uses parts of speech and phrases (NP, VP).
  • Syntactic function describes roles in a sentence (Subject, Object, Adverbial).
• Constituents
  • Words are organized into groupings that function as a whole.
  • Tested through linguistic tests of constituency.
• Phrase Structure Grammar (PSG)
  • Captures constituent status and ordering using context-free grammar.
  • Example rules: S \rightarrow NP VP, NP \rightarrow D N, VP \rightarrow V NP

Dependency Grammar (DG)

• An alternative to phrase structure.
• Syntactic functions are central.
• Syntactic structure consists of lexical items linked by binary asymmetric dependencies.
• Increasing interest in dependency-based parsing for NLP.
• Useful in relation extraction, question answering, and sentiment analysis.

Constituency vs. Relations

• DG is based on relationships between words (dependency relations).
• PSG is based on groupings or constituents.

Simple Relation Example

• In the sentence "The dog ate my homework", relations include:
  • ate →subj The dog
  • ate →obj my homework

Comparison

• Dependency structures represent head-dependent relations, functional categories, and parts-of-speech.
• Phrase structures represent phrases, structural categories, and grammatical functions.

Criteria for Heads and Dependents

1. H determines the syntactic category of C; H can replace C.
2. H determines the semantic category of C; D specifies H.
3. H is obligatory; D may be optional.
4. The form of D depends on H (agreement or government).
5. The linear position of D is specified with reference to H.

Some Tricky Cases

• Complex verb groups
• Subordinate clauses
• Coordination
• Prepositional phrases
• Punctuation

Dependency Graphs

• Defined as a directed graph G with:
  • A set V of nodes.
  • A set E of arcs (edges).
  • Labeled graphs with word forms and dependency types.
• Notations: i \rightarrow j \equiv (i, j) \in E

Formal Properties of Dependency Graphs

• antisymmetric: if A → B, then B ↛ A
• antireflexive: if A → B, then B ≠ A
• antitransitive: if A → B and B → C, then A ↛ C
• labeled: ∀ →, → has a label (r)

Formal Conditions on Dependency Graphs

• G is (weakly) connected: For every node i, there is a node j such that i → j or j → i.
• G is acyclic: If i → j then not j →∗ i.
• G obeys the single-head constraint: If i → j, then not k → j, for any k ≠ i.

Projectivity

• A projective graph: If i → j then for any k such that i< k < j or j < k < i, i →∗ k.
• Non-projective structures are needed for long-distance dependencies and free word order.

Treebanks

• Collections of sentences manually annotated with syntactic analysis.
• Used to train data-driven NLP tools.
• Examples: Penn Treebank, Prague Dependency Treebank, Negra/Tuba-DZ, Penn (Chinese), Norwegian Dependency Treebank, Universal Dependencies.

Norwegian Dependency Treebank (NDT)

• Completed in 2014 by Språkbanken, National Library.
• Ca 600,000 tokens of Bokmål and Nynorsk text.
• Enables training of taggers and parsers for Norwegian.
• Converted to Universal Dependencies.

Universal Dependencies

• Harmonized dependency treebanks for more than 100 languages.
• Norwegian models available in spaCy and Stanza.

CoNLL-U format

• Standard format for dependency treebanks.