260 likes | 280 Views
Between dependency structure and phrase structure . Dick Hudson UCL June 2018. How’s theoretical syntax doing?. Where are we coming from? Did it all start in 1957? Where are we now? Have we got the basics sorted? Are we converging on a single theory? Where would we like to be?
E N D
Between dependency structure and phrase structure Dick Hudson UCL June 2018
How’s theoretical syntax doing? • Where are we coming from? • Did it all start in 1957? • Where are we now? • Have we got the basics sorted? • Are we converging on a single theory? • Where would we like to be? • Will we still be here in 40 years?
My plan • I’ll give some historical context • Focusing on one issue: the nature of sentence structure. • Phrase structure (PS): • only phrases and their parts; no direct links between words • Dependency structure (DS): • only relations between words; no phrases • I’ll introduce Word Grammar • Relevant because it assumes DS • Started in 1984 but still evolving • I’ll admit to a crucial weakness in DS • But I’ll offer a solution. • Which is half-way between DS and PS.
For example For example S syntactic theory attracts good students VP o a s a NP NP J N V J N syntactic theory attracts good students JP JP dependency relation J N V J N whole-part relation
1. History of syntax: in a nutshell • The mainstream is a search for a single relation throughout the sentence. • Progress was gradual and erratic • bits of the wheel were rediscovered several times • The mainstream solution is DS (dependency structure) • The single relation includes: • adjuncts, complements, subjects • head = verb, noun, any word • PS is a historical aberration introduced by • a philosopher: Aristotle • a psychologist: Wundt
A > B = ‘B depends on A’ 1. History of syntax: the birth of grammar • -2,000 Babylon: word paradigms, no syntax • -500 India; Panini: DS (V > semantically defined kārakas) • but only applies to dependents of verbs. • -384 Greece; Aristotle: PS (proposition/sentence = subject/noun + predicate/verb) S V N
1. History of syntax: the Arabic grammarians • -2,000 Babylon: word paradigms, no syntax • -500 India; Panini: DS (A > B = ‘B depends on A’: V > semantically defined kārakas) • -384 Greece; Aristotle: PS (proposition = subject + predicate) • +350 Greece, Italy; Donatus, etc.: little syntax • +760 Baghdad; Sibawayh: DS (verb/prep > governed cases) • Basra and Kufa argued about grammatical theory: • Agreed: • Nouns and verbs have ‘case’ chosen by a ‘governor’. • This dependency also predicts word order. • Disputed: is mutual dependency possible? • Kufa: yes. • Basra: no.
1. History of syntax: Medieval Europe • +350 Greece, Italy; Donatus, etc.: little syntax • +760 Baghdad; Sibawayh: DS (verb/prep > governed cases) • +1150 France; Peter Helias: DS (verb/prep > governed case) • +1310 Germany; Thomas of Erfurt: DS (N > Adj, but subject/object N > V!) o s J N V J N
1. History of syntax: 18th century France • +1310 Germany; Thomas of Erfurt: DS (N > Adj, but subject/object N > V!) • 1765 France; Beauzée: DS+PS (N>…+V>…) • For evidence of the effect, see Google N-grams ‘analyse de la phrase’, ‘analyse des phrases’ • max 0.000008% = 8/100,000,000 • Already visible by 1800
1. History of syntax: 19th century Germany • +1310 Germany; Thomas of Erfurt: DS (N > Adj, but subject/object N > V!) • 1765 France; Beauzée: DS+PS (N>…+V>…) • +1826 Germany; Becker: DS+PS (first syntactic analysis table)
1. History of syntax: the first ‘tree’ • +1310 Germany; Thomas of Erfurt: DS (N > Adj, but subject/object N > V!) • +1765 France; Beauzée: DS+PS (N>…+V>…) • +1826 Germany; Becker: DS+PS (first syntactic analysis table) • +1834 Germany; Billroth: DS+PS (first syntactic diagram)
Google N-grams again: German ‘Satzbau’ • From 1820 • Peak is 10 times French peak (max = 9/10,000,000) • Apparently no influence from or on France.
1. History of syntax: verb-rooted trees • +1765 France; Beauzée: DS+PS (N>…+V>…) • +1826 Germany; Becker: DS+PS (first syntactic analysis table) • +1834 Germany; Billroth: DS+PS (first syntactic diagram) • +1873-84 Hungary, Russia, Germany: pure DS (verb > dependents > dependents) • Hungarian (Brassai) and German (Kern) also produced verb-centred tree diagrams. • NB 86 years before Tesnière (1959).
1. History of syntax: American trees • +1873-84 Hungary, Russia, Germany: pure DS (verb > dependents > dependents) • +1845 USA; Barrett: DS (first DS diagram, but rooted in the subject) • +1877 USA, Reed and Kellogg: DS+PS diagrams patented • Diagram by Sentence Diagrammer App!
Google N-grams: English ‘sentence structure’ • Not before 1880, but steep rise 1900-1940 – NOT due to linguistics! • But maybe this rise prepared a new generation of syntax-ready linguists? • No influence from Germany
1. History of syntax: Bloomfield’s Germany • +1877 USA; Reed and Kellogg: DS+PS diagrams patented • +1900 Germany; Wundt: PS (whole-part relations) • structure for: A sincerely thinking person scorns deception. G = ‘total meaning’ A = subject, B = predicate A1 = a person B1 = thinks sincerely A3 = thought B3 = is sincere A2 = deception B2 = is scorned
1. History of syntax: Bloomfield and followers head < dependent • +1877 USA; Reed and Kellogg: DS+PS diagrams patented • +1900 Germany; Wundt: PS (whole-part relations) • +1933 USA; Bloomfield: DS+PS (IC analysis = whole-part analysis ) • First trees in Nida 1943 x = exocentric (no head)
1. History of syntax: Chomsky S • +1933 USA; Bloomfield: DS+PS (IC analysis = whole-part analysis ) • +1957 USA; Chomsky: pure PS (survived only 13 years) • +1970 USA; Chomsky: PS with DS (X-bar syntax) • +1995 USA; Chomsky: DS? (Bare Phrase Structure) • No unary branching, so (3a) is replaced by (3b) NP VP This is a dependency. the book But what is this? the
2. Word Grammar • 1961: SOAS, then UCL: Halliday vs Chomsky • DS wasn’t mentioned. • 1964: PhD applying Halliday’s ‘Systemic Grammar’: very PS. • 1971: First formal grammar using Systemic Grammar. • 1976: ‘Daughter-Dependency Grammar’: PS + DS • 1984: ‘Word Grammar’: pure DS
The strengths of DS • A cognitive argument: • We can recognise relations between individual people, so why not between words? • A linguistic argument: • Words can select other words directly, e.g. • must + infinitive, but ought + to • give + to, present + with, bestow + upon • So we should at least allow DS in syntactic structure. • but if we have DS do we also need PS? • But simple DS is inadequate for the same reasons as simple PS.
For example: What did he stop doing? What did he stop doing? Mutual dependency! p c p s s x x,o x
But DS also needs extra nodes the book the • e.g. typical French houses needs distinct nodes for: • house – copied from the lexical entry • F-house – modified by French • t-F-house – modified by typical applied to ‘French house’ • pointed out by Oesten Dahl in 1980. • t-F-houses – affected by inflection: a set each of whose members is a typical French house. plural t-F-houses typical t-F-house inheritance French F-house house house But what is this? inheritance
The mystery relation: what is it? the isa • Not whole-part: • Nothing can be part of itself. • Not set-member: • Chomsky’s two the’s are both individuals, not sets. • Maybe: ‘isa’, the general-specific relation that carries inheritance. • book-the (the as combined with book) isathe • t-F-houses isa plural and t-F-house isaF-house isahouse • But: • ‘isa’ is the basis for grammatical competence • and these ‘sub-tokens’ are part of performance. book-the house house F-house French typical t-F-house plural t-F-houses
So competence meets performance word • Every token isa some stored type. • But a typical type isa some more general type. • And some tokens become permanent types. • So tokens are a transient fringe on the edge of the permanent grammar. • Moreover, tokens can isa each other • by grammatical modification • by deliberate repetition • by accidental repetition • by anaphora noun house plural house F-house t-F-house t-F-houses very very the Yes. the Is (it raining?)
So what? • Historically, DS is the mainstream. • Pure PS only lasted for 13 years. • Grammarians aren’t good at learning from the past and from other countries. • Maybe we’re all converging on DS? • If so, we need a combination of • DS between individual words • ‘isa’ between different analyses of the same word. • Maybe we can find a unified theory covering both competence and performance.
Thank you • This talk is available for download at http://dickhudson.com/talks/