Noun compounds NCs n Any sequence of nouns

Noun compounds (NCs) n Any sequence of nouns that itself functions as a noun n n asthma hospitalizations asthma hospitalization rates health care personnel hand wash Technical text is rich with NCs Open-labeled long-term study of the subcutaneous sumatriptan efficacy and tolerability in acute migraine treatment. 1

NCs: 3 computational tasks n n Identification Syntactic analysis (attachments) n n n [Baseline [headache frequency]] [[Tension headache] patient] Semantic analysis n n Headache treatment Corticosteroid treatment for headache treatment that uses corticosteroid 2

Two approaches n n Treat it as a classification problem (and use a machine learning algorithm) Linguistically motivated: consider the “semantics” of the nouns which will determine the relations between them 3

First approach Extraction of NCs from titles and abstracts of Medline n n n Part-of-Speech Tagger Extraction of sequences of units tagged as nouns Collection of 2245 NCs with 2 nouns A manual annotation of the NCs found 38 semantic relations Collection of labeled NCs and a set of semantic relations 4

Semantic relations n Frequency/time of n n Measure of n n headache drugs, hiv medications, influenza treatment Defect n n aciclovir therapy, laser irradiation, aerosol treatment “Purpose” n n relief rate, asthma mortality, hospital survival Instrument n n influenza season, headache interval hormone deficiency, csf fistulas, gene mutation Inhibitor n Adrenoreceptor blockers, influenza prevention 5

Semantic relations n Cause n n Change n n Bile delivery, virus reproduction Person Afflicted n n Papilloma growth, disease development Activity/Physical Process n n Asthma hospitalization, aids death …. Aids patients, headache group 6

Features n Lexical (words) n Me. SH descriptors 7

Classification method and results n n n Multi-class (18) classification problem Multi layer Neural Networks to classify across all relations simultaneously. Results Features Accuracy Words 62% Me. SH 61% Baselines Guessing 5% Most frequent relation 31% Vanderwende 94 (13 classes) 52% Lapata 00 (binary) 80% 8

Second approach n n Linguistic Motivation Head noun has argument structure n Meaning of the head noun determines what kinds of things can be done to it, what it is made of, what it is a part of… 9

Linguistic Motivation n Material + Cutlery Made of n n Food + Cutlery Used on n n steel knife, plastic fork, wooden spoon meat knife, dessert spoon, salad fork Profession + Cutlery Used by n chef's knife, butcher's knife 10

Linguistic Motivation n Hypothesis: n n A particular semantic relation holds between all 2 -word NCs that can be categorized by a Me. SH pair. Use the classes of Me. SH to identify semantic relations 11

Grouping the NCs n A 02 C 04 (Musculoskeletal System, Neoplasms) n n B 06 (Plants, Plants) n n skull tumors, bone cysts, bone metastases, skull osteosarcoma… eucalyptus trees, apple fruits, rice grains, potato plants A 01 M 01 (Body region, Person) n n n shoulder patient, eye physician, eye donor Too different: need to be more specific: go down the hierarchy A 01 M 01. 643 (Body Regions, Patients) n n shoulder patient C 04 M 01. 526 n (Body Regions, Occupational Groups) eye physician, chest physicians 12

Classification Decisions + Relations n n n A 02 C 04 Location of Disease B 06 Kind of Plants C 04 M 01 n C 04 M 01. 643 Person afflicted by Disease n C 04 M 01. 526 Person who treats Disease A 01 H 01 n A 01 H 01. 770 n A 01 H 01. 671. 538 n A 01 H 01. 671. 868 A 01 M 01 n A 01 M 01. 643 Person afflicted by Disease n A 01 M 01. 526 Specialist of n A 01 M 01. 898 Donor of 13

Evaluation n Accuracy: n n Anatomy: 91% accurate Natural Science: 79% Neoplasm: 100% Total Accuracy : 90. 8% 14

Conclusion of NCs n n n Problem of assigning semantic relations to two-word technical NCs Important problem: many NCs in technical text Especially difficult for the lack of syntactic clues State-of-the-art results One of very few working systems to tackle this task for NCs 15