Multiwords from Verb Complements Report

This task is to retrieve mulitwords from verb complements in the Lexicon. The Lexicon.2024 was to used for the development.. The Lexicon.2025 is the first public release. The growth report is shown as bellows..

• Light Verb Constructions
  • Performance on Candidates:

    Year	VC (Candidates)	LMWs (TP)	Not LMWs (FP)	Precision	Recall	F1
    2024	53	22	31	41.51	100.00	58.67
    2025	56	24	32	42.86	100.00	60.00

  • Performance of WordNet on candidates:

    Year	VC (Candidates)	Relevant		Irrelevant		Precision	Recall	F1
    		TP	FN	FP	TN
    2024	53	12	10	0	31	100.00	54.55	70.59
    2025	56	14	10	0	32	100.00	58.33	73.68

  • Summary:
    • The precision for multiwords from LVCs is low (42.86).
    • Relevant: 14 of the 24 multiword terms are multiwords in the WordNet (TP) with 10 are not in the WordNet (FN).
    • Irrelevant: All (32) not multiword terms in the Lexicon are not multiwords in the WordNet (TN) and none (0) not multiword terms are in the WordNet (FP).
    • WordNet multiword tagging has a precision of 100.
• Verb-Particle Constructions
  • Performance on Candidates:

    Year	VC (Candidates)	LMWs (TP)	Not LMWs (FP)	Precision	Recall	F1
    2024	2516	1682	834	66.85	100.00	80.13
    2025	2518	1684	834	66.88	100.00	80.15

  • Performance onf WordNet on candidates:

    Year	VC (Candidates)	Relevant		Irrelevant		Precision	Recall	F1
    		TP	FN	FP	TN
    2024	2516	1227	455	118	716	91.23	72.95	81.07
    2025	2518	1229	455	118	716	91.24	72.98	81.09

  • Summary:
    • The precision for multiwords from VPCs is decent (66.88).
    • Relevant: 1229 of the 1684 multiword terms are in the WordNet (TP) with 445 are not in the WordNet (FN).
    • Irrelevant: 716 of the not multiword term candidadtes (834) in the Lexicon are not multiwords in the WordNet (TN) and 118 not multiword term candidate are in the WordNet (FP).
    • WordNet multiword tagging has a high precision of 91.24 with 72.98 recall.
• Verb Complements (LVCs and VPCs)
  • Performance on Candidates:

    Year	VC (Candidates)	LMWs (TP)	Not LMWs (FP)	Precision	Recall	F1
    2024	2569	1704	865	66.33	100.00	79.76
    2025	2574	1708	866	66.36	100.00	79.78

  • Performance on WordNet on Candidate:

    Year	VC (Candidates)	Relevant		Irrelevant		Precision	Recall	F1
    		TP	FN	FP	TN
    2024	2569	1239	465	118	747	91.30	72.71	80.95
    2024	2574	1243	465	118	748	91.33	72.78	81.01

  • Summary on development based on Lexicon.2024:
    • The precision for multiword generation from VCs is decent (66.36).
    • 98.59% (= 1,684/1,708) of multiword are generated from the VPC model.
    • WordNet multiword tagging has a high precision of 91.33, recall of 72.78, and F1 of 80.01.

    • Relevant: 1243 of the 1708 multiword terms are in the WordNet (TP) with 465 are not in the WordNet (FN).
    • Irrelevant: 748 of the not multiword term candidadtes (866) from the Lexicon are not multiwords in the WordNet (TN) and 118 not multiword term candidate are in the WordNet (FP).