Understanding Qualia Structure in GL Theory for Semantic Problem-solving
Explore the Qualia structure in GL theory for semantic problem-solving, focusing on issues like polysemy and metonymy. Discover how label propagation aids in qualia classification, with examples ranging from noun compounds to verb-object relationships. Uncover the complexity of Qualia binding and its impact on entities across various contexts.
Download Presentation
Please find below an Image/Link to download the presentation.
The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author. Download presentation by click this link. If you encounter any issues during the download, it is possible that the publisher has removed the file from their server.
E N D
Presentation Transcript
Lunch talk Label propagation for qualia classification Tuan Do
Contents - Qualia structure in GL theory - Can we find GL representation? - Telic quale extraction from WN gloss for nouns. - Difficulty - Finding the link between entities instead - Qualia classification using label propagation
Qualia structure in GL theory - bake a cake bake cake EVENT = . ARGSTR = QUALIA = state_change_lcp ARGSTR = QUALIA = C = y F = x: food T = eat(e, z, x) A = bake_act (e, A = bake_act Arg1:animate, Arg2: (e,w,y) mass)
Qualia structure in GL theory - Qualia structure union bake cake EVENT = . ARGSTR = QUALIA = state_change_lcp ARGSTR = QUALIA = C = y F = x: food T = eat(e, z, x) A = bake_act (e, A = bake_act p:animate, y: (e,w,y) mass)
Qualia structure in GL theory - Qualia structure union bake a cake ARGSTR = QUALIA = F = exist(x) A = bake_act (e,p,y) T = eat(e, p, x)
Qualia structure in GL theory - Qualia binding happens between many type of entities - Verb object - Noun compound: horror film, gas mask, vacuum cleaner, silk dress - Modifier noun: useful knife, synthetic material - Subject object: The baker bake a cake
Qualia structure in GL theory - Qualia binding labeling vs semantic role labeling Agent Theme The singer sings a song Telic Formal Agentive
Qualia structure in GL theory - GL and qualia structure solve some interesting semantic problems. E.x. Polysemy: sense unification in Wordnet. - eat: take in solid food - eat: eat a meal - eat: take in food; used of animal only - eat: worry or cause anxiety - eat: cause to deteriorate
Qualia structure in GL theory - Metonymy: seize the throne seize throne EVENT = . ARGSTR = QUALIA = ARGSTR = QUALIA = F: x: chair T: T = control(y, z: sovereign) A = sit(y,x) / possess(y,x) A = possess(y,x)
Contents - Qualia structure in GL theory - Can we find GL representation? - Telic quale extraction from WN gloss for nouns. - Difficulty - Finding the link between entities instead - Qualia classification using label propagation
Telic quale extraction - Qualia extraction from wordnet gloss: aim at artifacts only. - Gloss example: - paint.n.01: a substance used as a coating to protect or decorate a surface - bed.n.01: a piece of furniture that provides a place to sleep
Telic quale extraction - Pattern matching: - for V-ing or S to V-inf - that S V / that V - 'used', 'intended', 'made', 'built', 'designed', in order to V - used in on V-ing - S is a NP to V-inf
Telic quale extraction - Qualia inheritance through Wordnet hierarchy: - hat - clothing: hat [protect - head - p_from weather]; clothing [be worn] hat QUALIA = F: x: clothing T: T = protect(z: y s head) A = wear(y,x) - It can only be semi-automatical
Contents - Qualia structure in GL theory - Can we find GL representation? - Telic quale extraction from WN gloss for nouns. - Difficulty - Finding the link between entities instead - Qualia classification using label propagation
GL representation difficulty - Is that easy to build GL representation resource? - Predicate is hard to keep consistent across the resource. - Should there be a fixed qualia structure or it is extensionable. - crafted cake - baked cake - make a cake
GL representation difficulty - Nested qualia structure - Predicate doesn t typically include preposition. E.g. MA driver MA driver QUALIA = QUALIA = F: x: F: x: human ? location T: drive(x, y: vehicle)
GL representation difficulty - Verb GL representation is hard to find. - Verb telic role is not typically shown in gloss. - Could be infered by associating nouns? - eat - food pair - Verb is idiosyncratic - Verbnet group might help?
Contents - Qualia structure in GL theory - Can we find GL representation? - Telic quale extraction from WN gloss for nouns. - Difficulty - Label qualia link between entities - Verb - object qualia link - Qualia link classification using label propagation
Verb-object qualia link - Acting on objects: - Typical formal verbs: Change-of-state verbs: changing in location, shape. - Typical telic verbs: some light verbs such as use , open (open a shop), prefer . - Both telic + formal: eat, drink, watch, listen, ride, shoot, fire, wear. For each verb, if the noun object is natural, it s more likely to act on formal quale.
Verb-object qualia link - Both agentive + formal: In most cases, it is just different senses of the same verb. - paint: - paint a painting (agentive), sense1 - paint the room (formal) sense2 - paint a person (formal) sense3 - bake: - bake a cake (agentive), sense2 - bake a potato (formal), sense1
Verb-object qualia link - Both consitutive + formal: meronymic intrinsic state change. Example: leak, repair, fix. It s better to collapse consitutive and formal into one class. No constitutive link found in a small annotated corpus of 40 documents. - -
Contents - Qualia structure in GL theory - Can we find GL representation? - Telic quale extraction from WN gloss for nouns. - Difficulty - Label qualia link between entities - Verb - object qualia link - Qualia link classification using label propagation
Label propagation - Semi-supervised learning algorithm: - Model: a network of nodes. Each node is a pair of verb - artifact noun - Some nodes are labelled, some nodes aren t - Label are propagated across the network by random walking algorithm
Label propagation - Similarity between nodes decide the transition probability. - Sim (verb1, noun1) - (verb2, noun2) = linear combination of sim(verb1, verb2), sim(noun1, noun2) - sim(verb1, verb2) = whether they re in same verb-net group - sim(noun1, noun2) = similarity of main sense in wordnet.
Label propagation - Have a closed form solution - solved by linear equations. - Transductive by nature but could be adapted to inductive learning. - Doesn t require too many seeding data. - Doesn t scale really well in training. - Unfortunately, I haven t gathered enough annotated data to test on.
Further investigations - Other type of entity pair: verb - head noun in prepositional phrase - Transitivity nature qualia link The singer sings a song Telic Telic Agentive Formal Agentive?
