Lexical Tools

Results of Proposed Rules - 2016

I. Results

11 SD-Rules are proposed to be added to the SD-Rule for evaluation. The results are described as follows:

Good/bad rules:
8 of them are evaluated as good rules in the optimized set
3 are bad rules.
Experts' suggestion: 50% (1/2) is good.
Computation rules: 78% (7/9) is good.
- NOM_D: 75% (3/4) is good
- ORG_D: 80% (4/5) is good
Also, in the optimized set, 2 child rules are used to replace proposed root-parent rules
- genesis$|noun|genic$|adj from esis$|noun|ic$|adj
- se$|verb|sis$|noun from e$|verb|is$|noun

SD-Rule	Rank	Precision	Instances	Source	Decompose	Results
Good Rules
`genesis$\|noun\|genic$\|adj`	13	99.52%	207	EXP_SUG	3G-Child	Good SD-Rule
`se$\|verb\|sis$\|noun`	27	97.87%	141	NOM_D	1G-Child	Good SD-Rule
`sia$\|noun\|tic$\|adj`	40	94.17%	103	ORG_D	Root-Parent	Good SD-Rule
`on$\|noun\|ve$\|adj`	48	91.46%	1253	ORG_D	Root-Parent	Good SD-Rule
`e$\|noun\|ic$\|adj`	49	91.40%	1267	ORG_D	Root-Parent	Good SD-Rule
`$\|adj\|ism$\|noun`	51	90.79%	369	NOM_D	Root-Parent	Good SD-Rule
`ation$\|noun\|ed$\|adj`	67	83.95%	405	NOM_D	Root-Parent	Good SD-Rule
`$\|noun\|ship$\|noun`	70	80.45%	133	ORG_D	Root-Parent	Good SD-Rule
Bad Rules
`e$\|adj\|ion$\|noun`	88	54.60%	359	NOM_D	Root-Parent	Bad SD-Rule
`$\|noun\|age$\|noun`	96	36.97%	119	ORG_D	Root-Parent	Bad SD-Rule
`al$\|adj\|ine$\|noun`	98	32.65%	49	EXP_SUG	Root-Parent	Bad SD-Rule

II. Further Observation on NOM_D

The top 22 SD-Rule generated from NOM_D are evaluated, 95.5% (21/22) are good SD-Rules.

Evaluated in 2015: Frequency > 200, Instance coverage > 1.00% , Accum. Coverage > 80.0%
ID	SD-Rule	Rank	Precision	Instances	Notes
1	`$\|adj\|ness$\|noun`	1	100.00%	2735	2013
2	`se$\|verb\|zation$\|noun`	2	100.00%	1109	2015
3	`sation$\|noun\|ze$\|verb`	3	100.00%	1072	2015
4	`ility$\|noun\|le$\|adj`	9	99.94%	1629	2015
5	`iness$\|noun\|y$\|adj`	10	99.82%	546	2013
6	`ation$\|noun\|e$\|verb`	15	99.25%	2519	2013
7	`ce$\|noun\|t$\|adj`	19	98.83%	854	2015, `nce$\|noun\|nt$\|adj` is used
8	`cy$\|noun\|t$\|adj`	20	98.77%	407	2015
9	`e$\|verb\|ion$\|noun`	21	98.76%	2337	2015
10	`$\|verb\|ment$\|noun`	22	98.35%	607	2013
11	`ication$\|noun\|y$\|verb`	23	98.33%	300	2013
12	`ed$\|adj\|ion$\|noun`	24	98.15%	704	2013
13	`$\|adj\|ity$\|noun`	26	97.89%	2082	2013
14	`e$\|adj\|ity$\|noun`	31	97.55%	897	2013
15	`$\|verb\|ion$\|noun`	41	93.72%	573	2013
16	`$\|verb\|ing$\|noun`	44	92.56%	524	2013
17	`$\|verb\|ation$\|noun`	52	90.29%	340	2013
Evaluated in 2016: Frequency > 100, Instance coverage > 0.40% , Accum. Coverage > 83.36%)
18	`e$\|verb\|ing$\|noun`	47	91.47%	211	2015, Expert's Suggested
19	`ation$\|noun\|ed$\|adj`	67	83.95%	405	2016
20	`$\|adj\|ism$\|noun`	51	90.79%	369	2016
21	`e$\|adj\|ion$\|noun`	88	54.60%	359	2016, bad (cutoff at Rank of 82)
22	`e$\|verb\|is$\|noun`	27	97.87%	141	2016, `se$\|verb\|sis$\|noun` is used

III. Further Observation on ORG_D

The top 6 SD-Rules generated from ORG_D are all good SD-Rules

Evaluated in 2015: Frequency > 40, Instance coverage > 1.00% , Accum. Coverage > 11.50%
ID	SD-Rule	Rank	Precision	Instances	status
1	`$\|noun\|less$\|adj`	11	99.64%	561	2013
2	`$\|verb\|ion$\|noun`	41	93.72%	573	2013, also from NOM_D
3	`ist$\|noun\|y$\|noun`	37	95.48%	509	2013
4	`$\|adj\|ally$\|adv`	17	99.08%	2072	2015
5	`$\|noun\|ful$\|adj`	54	89.21%	139	2013
6	`c$\|adj\|s$\|noun`	55	89.04%	292	2015
Evaluated in 2016: frequency >= 35; Accu. coverage: > 16.00% Ind Coverage: > 0.80%
7	`on$\|noun\|ve$\|adj`	48	91.46%	1253	2016
8	`$\|noun\|ship$\|noun`	70	80.45%	133	2016
9	`$\|noun\|age$\|noun`	96	36.97%	119	2016, bad (cutoff at Rank of 82)
10	`e$\|noun\|ic$\|adj`	49	91.40%	1267	2016
11	`sia$\|noun\|tic$\|adj`	40	94.17%	103	2016

V. Future Work

Evaluated more SD-Rules from NOM_D and ORG_D down the list.