University of Oslo : Department of Informatics A Syntactic Constituent Ranker for Speculation and Negation Scope Resolution Erik Velldal Jonathon Read† † Date: 21 September 2011 Lilja Øvrelid Stephan Oepen [email protected] Venue: Master Seminar in Language Technology Department of Informatics University of Oslo Introduction Task Given an appropriate cue of speculation or negation, determine its scope—the subsequence of the sentence that is affected. For example: I {The unknown amino acid may be used by these species}. I Samples of the protein pair space were taken {hinstead ofi considering the whole space} [...] Some applications: I Information Extraction I Sentiment Analysis Introduction Our approach Generate candidate scopes from constituents in syntactic trees; use a supervised learner to rank these candidates speculation / negation cue unlabeled sentence syntactic constituent parsing datadriven ranking predicted scope Outline 1 Data sets and evaluation measures 2 A brief review of speculation and negation at LNS 3 A syntactic constituent ranker 4 Experiments Outline 1 Data sets and evaluation measures 2 A brief review of speculation and negation at LNS 3 A syntactic constituent ranker 4 Experiments Data Sets BioScope Corpus (Vincze et al., 2008) I 20,924 sentences annotated with speculation/negation cues and associated scopes. I Sentences are drawn from biomedical abstracts, full papers and clinical reports. CoNLL-2010 Shared Task Evaluation Data (Farkas et al., 2010) I A further set of 5,003 sentences from biomedical papers, annotated for speculation. Data Sets Data Total Sentences Abstracts Papers CoNLL 11,871 2,670 5,003 Speculation Sentences Cues 2,101 519 790 2,659 668 1,033 Negation Sentences Cues 1,597 339 – 1,719 376 – Note on data set splits for development and evaluation: Speculation: all of the Abstracts and Papers are used for development; the CoNLL set is held-out. Negation: 90% of the Abstracts and Papers are used for development; the remainder is held-out. Preprocessing I BioScope XML converted to stand-off characterisation. I Tokenisation performed using the GENIA tagger (Tsuruoka et al., 2005), supplemented with a finite-state tokeniser I Part-of-speech tags from both the GENIA tagger (for higher accuracy in the biomedical domain) and TnT (Brants, 2000) I Dependency parsing using MaltParser (Nivre et al., 2006) stacked with the XLE platform (Crouch et al., 2008) with the English grammar developed by Butt et al., (2002). I Constituent tree parsing with PET (Callmeier, 2002) and the English Resource Grammar (Flickinger 2002). Preprocessing I BioScope XML converted to stand-off characterisation. I Tokenisation performed using the GENIA tagger (Tsuruoka et al., 2005), supplemented with a finite-state tokeniser I Part-of-speech tags from both the GENIA tagger (for higher accuracy in the biomedical domain) and TnT (Brants, 2000) I Dependency parsing using MaltParser (Nivre et al., 2006) stacked with the XLE platform (Crouch et al., 2008) with the English grammar developed by Butt et al., (2002). I Constituent tree parsing with PET (Callmeier, 2002) and the English Resource Grammar (Flickinger 2002). Preprocessing I BioScope XML converted to stand-off characterisation. I Tokenisation performed using the GENIA tagger (Tsuruoka et al., 2005), supplemented with a finite-state tokeniser I Part-of-speech tags from both the GENIA tagger (for higher accuracy in the biomedical domain) and TnT (Brants, 2000) I Dependency parsing using MaltParser (Nivre et al., 2006) stacked with the XLE platform (Crouch et al., 2008) with the English grammar developed by Butt et al., (2002). I Constituent tree parsing with PET (Callmeier, 2002) and the English Resource Grammar (Flickinger 2002). Preprocessing I BioScope XML converted to stand-off characterisation. I Tokenisation performed using the GENIA tagger (Tsuruoka et al., 2005), supplemented with a finite-state tokeniser I Part-of-speech tags from both the GENIA tagger (for higher accuracy in the biomedical domain) and TnT (Brants, 2000) I Dependency parsing using MaltParser (Nivre et al., 2006) stacked with the XLE platform (Crouch et al., 2008) with the English grammar developed by Butt et al., (2002). I Constituent tree parsing with PET (Callmeier, 2002) and the English Resource Grammar (Flickinger 2002). Preprocessing I BioScope XML converted to stand-off characterisation. I Tokenisation performed using the GENIA tagger (Tsuruoka et al., 2005), supplemented with a finite-state tokeniser I Part-of-speech tags from both the GENIA tagger (for higher accuracy in the biomedical domain) and TnT (Brants, 2000) I Dependency parsing using MaltParser (Nivre et al., 2006) stacked with the XLE platform (Crouch et al., 2008) with the English grammar developed by Butt et al., (2002). I Constituent tree parsing with PET (Callmeier, 2002) and the English Resource Grammar (Flickinger 2002). Evaluation measures Evaluation was performed using the scoring software of the CoNLL 2010 Shared Task. Prec = tp tp+fp Rec = tp tp+fn F1 = 2×Prec×Rec Prec+Rec A true positive requires identification of all words in scope. Outline 1 Data sets and evaluation measures 2 A brief review of speculation and negation at LNS 3 A syntactic constituent ranker 4 Experiments Supervised classification of cues We employ a binary word-by-word linear support vector machine classifier using the SVMlight toolkit. A simpler approach is to treat the set of cues as near-closed class, applying filtering to disregard all known non-cues before training and applying the classifier. Cues described using features of: I n-grams of lemmas up to three positions to the left; and I n-grams of surface forms up to two positions to the right. Supervised classification of cues We employ a binary word-by-word linear support vector machine classifier using the SVMlight toolkit. A simpler approach is to treat the set of cues as near-closed class, applying filtering to disregard all known non-cues before training and applying the classifier. Cues described using features of: I n-grams of lemmas up to three positions to the left; and I n-grams of surface forms up to two positions to the right. Supervised classification of cues We employ a binary word-by-word linear support vector machine classifier using the SVMlight toolkit. A simpler approach is to treat the set of cues as near-closed class, applying filtering to disregard all known non-cues before training and applying the classifier. Cues described using features of: I n-grams of lemmas up to three positions to the left; and I n-grams of surface forms up to two positions to the right. Supervised classification of cues Multi-word cues are handled in post-processing by matching patterns observed in the development data lemmas. Speculation cannot {be}? exclude either .+ or indicate that may,? or may not no {evidence | proof | guarantee} not {known | clear | evident | understood | exclude} raise the .* {possibility | question | issue | hypothesis} whether or not Negation rather than {can|could} not no longer instead of with the * exception of neither * nor {no(t?)|neither} * nor Supervised classification of cues Development Results: Speculation Baseline Word-by-word Filtering Prec Rec F1 90.49 94.65 94.13 81.16 82.26 84.60 85.57 88.02 89.11 Prec Rec F1 87.69 92.49 97.61 97.72 92.38 95.03 Development Results: Negation Word-by-word Filtering Supervised classification of cues Development Results: Speculation Baseline Word-by-word Filtering Prec Rec F1 90.49 94.65 94.13 81.16 82.26 84.60 85.57 88.02 89.11 Prec Rec F1 87.69 92.49 97.61 97.72 92.38 95.03 Development Results: Negation Word-by-word Filtering Hueristics for scope resolution Hueristics, activated by the cue’s part-of-speech operate over dependency structures: I Coordinations scope over their conjuncts; I Prepositions scope over their argument and its descendants; I Attributive adjectives scope over their nominal head and its descendants; I Predicative adjectives scope over referential subjects and clausal arguments, if present; I Modals inherit subject scope from their lexical verb and scope over their descendants; and I Passive or raising verbs scope over referential subjects and the verbal descendants. Hueristics for scope resolution Hueristics, activated by the cue’s part-of-speech operate over dependency structures: I Coordinations scope over their conjuncts; I Prepositions scope over their argument and its descendants; I Attributive adjectives scope over their nominal head and its descendants; I Predicative adjectives scope over referential subjects and clausal arguments, if present; I Modals inherit subject scope from their lexical verb and scope over their descendants; and I Passive or raising verbs scope over referential subjects and the verbal descendants. Hueristics for scope resolution Hueristics, activated by the cue’s part-of-speech operate over dependency structures: I Coordinations scope over their conjuncts; I Prepositions scope over their argument and its descendants; I Attributive adjectives scope over their nominal head and its descendants; I Predicative adjectives scope over referential subjects and clausal arguments, if present; I Modals inherit subject scope from their lexical verb and scope over their descendants; and I Passive or raising verbs scope over referential subjects and the verbal descendants. Hueristics for scope resolution Hueristics, activated by the cue’s part-of-speech operate over dependency structures: I Coordinations scope over their conjuncts; I Prepositions scope over their argument and its descendants; I Attributive adjectives scope over their nominal head and its descendants; I Predicative adjectives scope over referential subjects and clausal arguments, if present; I Modals inherit subject scope from their lexical verb and scope over their descendants; and I Passive or raising verbs scope over referential subjects and the verbal descendants. Hueristics for scope resolution Hueristics, activated by the cue’s part-of-speech operate over dependency structures: I Coordinations scope over their conjuncts; I Prepositions scope over their argument and its descendants; I Attributive adjectives scope over their nominal head and its descendants; I Predicative adjectives scope over referential subjects and clausal arguments, if present; I Modals inherit subject scope from their lexical verb and scope over their descendants; and I Passive or raising verbs scope over referential subjects and the verbal descendants. Hueristics for scope resolution Hueristics, activated by the cue’s part-of-speech operate over dependency structures: I Coordinations scope over their conjuncts; I Prepositions scope over their argument and its descendants; I Attributive adjectives scope over their nominal head and its descendants; I Predicative adjectives scope over referential subjects and clausal arguments, if present; I Modals inherit subject scope from their lexical verb and scope over their descendants; and I Passive or raising verbs scope over referential subjects and the verbal descendants. Hueristics for scope resolution Hueristics, activated by the cue’s part-of-speech operate over dependency structures: I Coordinations scope over their conjuncts; I Prepositions scope over their argument and its descendants; I Attributive adjectives scope over their nominal head and its descendants; I Predicative adjectives scope over referential subjects and clausal arguments, if present; I Modals inherit subject scope from their lexical verb and scope over their descendants; and I Passive or raising verbs scope over referential subjects and the verbal descendants. Hueristics for scope resolution In the case of negation an additional set of rules are applied: I Determiners scope over their head node and its descendants; I The noun none scopes over the entire sentence if it is the subject, otherwise it scopes over its descendants; I Other nouns or verbs scope over their descendants; I Adverbs with a verbal head scope over the descendants of the lexical verb; and I Other adverbs scope over the descendants of the head. If none of these rules activate then a default scope is applied by labeling from the cue to the sentence final punctuation. Hueristics for scope resolution In the case of negation an additional set of rules are applied: I Determiners scope over their head node and its descendants; I The noun none scopes over the entire sentence if it is the subject, otherwise it scopes over its descendants; I Other nouns or verbs scope over their descendants; I Adverbs with a verbal head scope over the descendants of the lexical verb; and I Other adverbs scope over the descendants of the head. If none of these rules activate then a default scope is applied by labeling from the cue to the sentence final punctuation. Hueristics for scope resolution In the case of negation an additional set of rules are applied: I Determiners scope over their head node and its descendants; I The noun none scopes over the entire sentence if it is the subject, otherwise it scopes over its descendants; I Other nouns or verbs scope over their descendants; I Adverbs with a verbal head scope over the descendants of the lexical verb; and I Other adverbs scope over the descendants of the head. If none of these rules activate then a default scope is applied by labeling from the cue to the sentence final punctuation. Hueristics for scope resolution In the case of negation an additional set of rules are applied: I Determiners scope over their head node and its descendants; I The noun none scopes over the entire sentence if it is the subject, otherwise it scopes over its descendants; I Other nouns or verbs scope over their descendants; I Adverbs with a verbal head scope over the descendants of the lexical verb; and I Other adverbs scope over the descendants of the head. If none of these rules activate then a default scope is applied by labeling from the cue to the sentence final punctuation. Hueristics for scope resolution In the case of negation an additional set of rules are applied: I Determiners scope over their head node and its descendants; I The noun none scopes over the entire sentence if it is the subject, otherwise it scopes over its descendants; I Other nouns or verbs scope over their descendants; I Adverbs with a verbal head scope over the descendants of the lexical verb; and I Other adverbs scope over the descendants of the head. If none of these rules activate then a default scope is applied by labeling from the cue to the sentence final punctuation. Hueristics for scope resolution In the case of negation an additional set of rules are applied: I Determiners scope over their head node and its descendants; I The noun none scopes over the entire sentence if it is the subject, otherwise it scopes over its descendants; I Other nouns or verbs scope over their descendants; I Adverbs with a verbal head scope over the descendants of the lexical verb; and I Other adverbs scope over the descendants of the head. If none of these rules activate then a default scope is applied by labeling from the cue to the sentence final punctuation. Hueristics for scope resolution In the case of negation an additional set of rules are applied: I Determiners scope over their head node and its descendants; I The noun none scopes over the entire sentence if it is the subject, otherwise it scopes over its descendants; I Other nouns or verbs scope over their descendants; I Adverbs with a verbal head scope over the descendants of the lexical verb; and I Other adverbs scope over the descendants of the head. If none of these rules activate then a default scope is applied by labeling from the cue to the sentence final punctuation. Hueristics for scope resolution Development Results: Speculation Scope for Gold Cues Abstracts Default Baseline Dependency Rules F1 69.84 73.67 Papers Default Baseline Dependency Rules 45.21 72.31 Development Results: Negation Scope for Gold Cues Abstracts Default Baseline Dependency Rules F1 51.75 70.76 Papers Default Baseline Dependency Rules 31.38 67.16 Hueristics for scope resolution Development Results: Speculation Scope for Gold Cues Abstracts Default Baseline Dependency Rules F1 69.84 73.67 Papers Default Baseline Dependency Rules 45.21 72.31 Development Results: Negation Scope for Gold Cues Abstracts Default Baseline Dependency Rules F1 51.75 70.76 Papers Default Baseline Dependency Rules 31.38 67.16 Outline 1 Data sets and evaluation measures 2 A brief review of speculation and negation at LNS 3 A syntactic constituent ranker 4 Experiments Selecting candidate constituents Learn a ranking function over candidate constituents within a parse. sb-hd mc c sp-hd n c d - the le the hd-cmp u c aj-hdn norm c aj - i le n ms-cnt ilr unknown n - mc le amino acid hd-cmp u c v vp mdl-p le may v prd be le hd-cmp u c be v pas odlr v np le hd-cmp u c p np ptcl le sp-hd n c used by d - pl le w period plr these n pl olr n - c le species. Selecting candidate constituents Candidates are generated by following the path from the cue to the root of the tree, for example: I modal verb : False The unknown amino acid { may } by used by these species. I head – complement : False The unknown amino acid { may be used by these species}. I subject – head : True {The unknown amino acid may be used by these species}. Selecting candidate constituents Candidates are generated by following the path from the cue to the root of the tree, for example: I modal verb : False The unknown amino acid { may } by used by these species. I head – complement : False The unknown amino acid { may be used by these species}. I subject – head : True {The unknown amino acid may be used by these species}. Selecting candidate constituents Candidates are generated by following the path from the cue to the root of the tree, for example: I modal verb : False The unknown amino acid { may } by used by these species. I head – complement : False The unknown amino acid { may be used by these species}. I subject – head : True {The unknown amino acid may be used by these species}. Alignment of constituents and scopes Scope boundaries must align with the boundaries of constituents for the approach to be successful. Alignment can be improved by applying slackening rules, including: I eliminating constituent-final punctuation; I eliminating constituent-final parenthesised elements; I reducing scope to the left when the left-most terminal is an adverb and not the cue; and I ensuring the scope starts with the cue when it is a noun. Alignment of constituents and scopes Scope boundaries must align with the boundaries of constituents for the approach to be successful. Alignment can be improved by applying slackening rules, including: I eliminating constituent-final punctuation; I eliminating constituent-final parenthesised elements; I reducing scope to the left when the left-most terminal is an adverb and not the cue; and I ensuring the scope starts with the cue when it is a noun. % of hedge scopes aligned with constituents Alignment of constituents and scopes 95 90 85 BSA BSP 80 1 10 20 30 n-best parsing results 40 50 Alignment of constituents and scopes Analysis of the non-aligned items indicated that mismatches arise from: I parse ranking errors (40%), I non-syntactic scope (25%) e.g. “This allows us to { address a number of questions : what proportion of [...] can be attributed to a known domain-domain interaction}?”, I divergent syntactic theories (16%), I parenthesised elements (13%) and I annotation errors (6%) Alignment of constituents and scopes In the papers: I A parse is produced for 86% of sentences. I Alignment of constituents and speculation scopes is 81% when inspecting the first parse of sentences. I The upper-bound performance of the ranker is around 76% when resolving the scope of speculation. I A similar analysis indicates an upper-bound of 77% for negation. Alignment of constituents and scopes In the papers: I A parse is produced for 86% of sentences. I Alignment of constituents and speculation scopes is 81% when inspecting the first parse of sentences. I The upper-bound performance of the ranker is around 76% when resolving the scope of speculation. I A similar analysis indicates an upper-bound of 77% for negation. Training procedure Given a parsed BioScope sentence: I The candidate consitutuent that corresponds to the scope is labeled as correct; I all other candidates are labeled as incorrect. I Learn a scoring function (using SVMlight ). Describe candidates with three classes of features that record: I paths in the tree; I surface order; and I rule-like linguistic phenomena. Training procedure sb-hd mc c sp-hd n c d - the le the hd-cmp u c aj-hdn norm c aj - i le n ms-cnt ilr unknown n - mc le amino acid hd-cmp u c v vp mdl-p le may v prd be le hd-cmp u c be v pas odlr v np le hd-cmp u c p np ptcl le sp-hd n c used by d - pl le w period plr these n pl olr n - c le species. Path features Paths from speculation cues to candidate constituents: I specific, e.g. ‘‘v vp mdl-p le\hd-cmp u c\sb-hd mc c’’ I general, e.g. ‘‘v vp mdl-p le\\sb-hd mc c’’ With lexicalised variants: I specific, e.g. ‘‘may : v vp mdl-p le\hd-cmp u c\sb-hd mc c’’ I general, e.g. ‘‘may : v vp mdl-p le\\sb-hd mc c’’ Bigrams formed of nodes and their parents, e.g. ‘‘v vp mdl-p le hd-cmp u c’’ and ‘‘hd-cmp u c sb-hd mc c’’ Surface features I Bigrams formed of preterminal lexical types, e.g. ‘‘d - the le aj-hdn norm c’’, ‘‘aj-hdn norm c n - mc le’’, etc. I Cue position within candidate (in tertiles) , e.g. ‘‘33%-66%’’. I Candidate size relative to sentence length (in quartiles), e.g. ‘‘75%-100%’’ I Punctuation preceeding the candidate, e.g. ‘‘false’’ I Punctuation at end of the candidate e.g. ‘‘true’’ Rule-like features Detection of: I Passivisation I Subject control verbs occuring with passivised verbs I Subject raising verbs I Predicative adjectives Detecting control verbs with a passivised verb 1. Cue is a subject control verb. subject– head 2. Find the first subject – head parent of (1) on the head path. (2) 3. Find the first head – complement child of (2) on the head path. 4. The right-most daughter of (3) or one of its descendents is a passivized verb. 5. The transitive head daughter of the left-most daughter of (2) is not an expletive it or there. H * H (3) head– complement (5) H not expletive pronoun subject (4) control passive verb verb (1) the unknown amino acid <may> be used by these species. Using predictions from auxiliary systems Combine with other systems by introducing features recording matches with: I the start of the auxiliary prediction; I the end of the auxiliary prediction; or I the entirety of the auxiliary prediction. In cases where an ERG parse is unavailable we back off to the prediction of the dependency rules. Outline 1 Data sets and evaluation measures 2 A brief review of speculation and negation at LNS 3 A syntactic constituent ranker 4 Experiments Ranker optimisation Speculation and negation in aligned items from papers (F1) Random baseline Path Path+Surface Path+Rule-like Path+Surface+Rule-like Speculation Negation 26.76 78.10 79.93 83.72 85.30 21.73 79.28 78.88 89.47 89.24 Training with the first aligned constituent in n-best parses and testing with m-best parses did not greatly impact performance, but optimal values are: n = 1 and m = 3. Ranker optimisation Speculation and negation in aligned items from papers (F1) Random baseline Path Path+Surface Path+Rule-like Path+Surface+Rule-like Speculation Negation 26.76 78.10 79.93 83.72 85.30 21.73 79.28 78.88 89.47 89.24 Training with the first aligned constituent in n-best parses and testing with m-best parses did not greatly impact performance, but optimal values are: n = 1 and m = 3. n-best optimisation for combined system F1 77 76 75 25 74 0 20 5 10 training n-best15 20 5 25 0 15 10 testing m-best Development results Resolving scope of gold cues (F1) Default baseline Constituent ranker Dependency rules Combined Speculation Negation 64.89 73.67 73.39 78.69 48.07 67.46 70.11 73.98 Evaluation results We evaluate end-to-end performance for speculation on the CoNLL-2010 Shared Task evaluation data. Speculation System Prec Rec F1 Morante et al. (2010) Cue classifier + combined 59.62 62.00 55.18 57.02 57.32 59.41 Evaluation results Negation (cross-validated on abstracts) System Prec Rec F1 Morante et al. (2009) Cue classifier + combined 66.31 68.93 65.27 73.03 65.79 70.92 System Prec Rec F1 Morante et al. (2009) Cue classifier + combined 42.49 61.77 39.10 71.55 40.72 66.30 Negation (on held-out papers) Evaluation results Negation (cross-validated on abstracts) System Prec Rec F1 Morante et al. (2009) Cue classifier + combined 66.31 68.93 65.27 73.03 65.79 70.92 System Prec Rec F1 Morante et al. (2009) Cue classifier + combined 42.49 61.77 39.10 71.55 40.72 66.30 Negation (on held-out papers) Conclusions I It is possible to learn a discriminative ranking function for choosing subtrees from HPSG-based constituent structures that match speculation scopes; I Combining the rule-based and ranker-based approaches to resolve the scope of cues predicted by our classifier achieves the best results to date on the CoNLL 2010 Shared Task evaluation data; I The system is readily adapted to deal with negation, also achieving state-of-the-art results. Conclusions I It is possible to learn a discriminative ranking function for choosing subtrees from HPSG-based constituent structures that match speculation scopes; I Combining the rule-based and ranker-based approaches to resolve the scope of cues predicted by our classifier achieves the best results to date on the CoNLL 2010 Shared Task evaluation data; I The system is readily adapted to deal with negation, also achieving state-of-the-art results. Conclusions I It is possible to learn a discriminative ranking function for choosing subtrees from HPSG-based constituent structures that match speculation scopes; I Combining the rule-based and ranker-based approaches to resolve the scope of cues predicted by our classifier achieves the best results to date on the CoNLL 2010 Shared Task evaluation data; I The system is readily adapted to deal with negation, also achieving state-of-the-art results.
© Copyright 2026 Paperzz