Have a personal or library account? Click to login
A New Corpus of Lexical Substitution and Word Blend Errors: Probing the Semantic Structure of Lemma Access Failures Cover

A New Corpus of Lexical Substitution and Word Blend Errors: Probing the Semantic Structure of Lemma Access Failures

Open Access
|May 2023

Figures & Tables

Table 1

Relative frequencies of word errors by semantic category (columns) and contextuality, form-relatedness, correctedness, and error type (rows). Percentages in parentheses are by row totals, except for correctedness, which is by column total.

CO-HYPONYMSANTONYMSSYNONYMSSUBSUMATIVESTHEMATICUNRELATEDTOTALS
Contextual23 (10.75)5 (2.34)6 (2.80)10 (4.67)52 (24.30)118 (55.14)214
Noncontextual (NC)222 (25.23)41 (4.66)77 (8.75)61 (6.93)153 (17.39)326 (37.05)880
Form-related18 (7.73)5 (2.15)17 (7.30)4 (1.72)13 (5.58)176 (75.54)233
Not form-related (NFR)227 (26.36)41 (4.76)66 (7.67)67 (7.78)192 (22.30)268 (31.13)861
Corrected215 (87.76)37 (80.43)66 (79.52)63 (88.73)162 (79.02)335 (75.45)878
Not corrected30 (12.24)9 (19.57)17 (20.48)8 (11.27)43 (20.98)109 (24.55)216
Word blends23 (21.70)4 (3.77)38 (35.85)5 (4.72)15 (14.15)21 (19.81)106
Lexical substitutions245 (22.39)46 (4.20)83 (7.59)71 (6.49)205 (18.74)444 (40.59)1094
NC and NFR206 (30.70)36 (5.37)61 (9.09)57 (8.49)144 (21.46)167 (24.89)671
Table 2

Synonym and subsumative errors (with /intended/→error, SFUSED-English record ID number, and podcast source).

A. Synonyms: Same concept, ungrammatical by selectional restrictions: 16 (19.28%)
  1. What did we /say my, what did we tell my parents we’re seeing? (/tell/→say, 2783, Battleship Pretension)

  2. I can see how one is totally /badder, worse than the other. (/worse/→badder, 205, Go Bayside)

  3. We can only see a /few percentage of the whole universe. (/small/→few, 2358, Astronomy Cast)

B. Synonyms: Same/close concept, but odd: 26 (31.33%)
  1. Don’t bring those kids in the /home, house. (/house/→home, 9543, This Feels Terrible)

  2. Professional /level, uh professional quality ear buds, yes. (/quality/→level, 6688, Battleship Pretension)

  3. I didn’t know this at the /point, time that they were dating. (/time/→point, 8337, This Feels Terrible)

C. Synonyms: same/close concept, but not right: 41 (49.40%)
  1. And if we don’t understand gravity at those /long [d]= large distances then. (/large/→ long, 4593, Astronomy Cast)

  2. We’ve /dedicated, not dedicated, devoted a whole episode to him. (/devoted/→dedicated, 4557, Battleship Pretension)

  3. The /Dark Ages, the Middle Ages were stupid. (/Middle Ages/→Dark Ages, 3224, We Have Concerns)

D. Subsumatives: super-ordinates 47 (66.20%)
  1. So maybe it was too nitrogen rich and wasn’t putting out /plants, er, not plants, flowers. (/flowers/→plants, 1739, direct observation)

  2. … about the feelings of the /people that she, the men that she loves (/men/→people, 5238, Battleship Pretension)

  3. If there’s one thing to with /anger, rage that’s healthy, it’s to dance it out. (/rage/→anger, 5361, We Have Concerns)

E. Subsumatives: subordinates: 24 (33.80%)
  1. Is there anything else you wanted to say about that /song, or that, that music? (/music/→song, 7248, Battleship Pretension)

  2. You have a complex /water cycle, or liquid cycle rather. (/liquid/→water, 9906, Astronomy Cast)

  3. I like melted /congee. (/rice/→congee, 4359, direct observation)

Table 3

Total counts by category (percentage occurrence), mean detection rates (i.e., counts/hours of listening), and 95% confidence interval for estimated Poisson rates.

CO-HYPONYMSANTONYMSSYNONYMSSUBSUMATIVESTHEMATICUNRELATED
Total (%)194 (23.69)34 (4.15)70 (8.55)57 (6.96)158 (19.29)306 (37.36)
Detection rates0.830.120.270.220.471.09
CI 95% Poisson221.30, 166.7047.51, 23.5588.44, 54.5773.85, 43.17182.64, 133.36340.29, 271.71
joc-6-1-278-g1.png
Figure 1

Estimates of Poisson Rates with 95% CI.

DOI: https://doi.org/10.5334/joc.278 | Journal eISSN: 2514-4820
Language: English
Submitted on: Mar 2, 2023
Accepted on: Apr 14, 2023
Published on: May 18, 2023
Published by: Ubiquity Press
In partnership with: Paradigm Publishing Services
Publication frequency: 1 issue per year

© 2023 John Alderete, Melissa Baese-Berk, Adrian Brasoveanu, Jess H. K. Law, published by Ubiquity Press
This work is licensed under the Creative Commons Attribution 4.0 License.