4 Matching Individual Items

Sometimes you might want to hand-pick items from a list of candidates. For example, you might want a matched word to be a plausible replacement for a target word in a sentence. This is possible with match_word().

4.1 Example

Here’s an example usage of match_item(), to suggest a word matched for “elephant” in terms of:

• Length (exactly)
• Frequency (within ±0.25 Zipf)
• Imageability (within ±1, on a 1-7 Likert rating scale)
• Part of Speech (i.e. is also a noun)

library(LexOPS)

suggested_matches <- lexops |>
  match_item(
    "elephant",
    Length,
    Zipf.SUBTLEX_UK = -0.25:0.25,
    IMAG.Glasgow_Norms = -1:1,
    PoS.SUBTLEX_UK
  )

The suggested matches are returned in a dataframe, filtered to be within the specified tolerances, and ordered by euclidean distance from the target word (calculated using all the numeric variables used). The closest suggested match for “elephant” is “sandwich”. If we are looking for a match to fit in a sentential context, we can choose the best suitable match from this list. (For space, only the top 10 matches are shown).

string	euclidean_distance	Length	Zipf.SUBTLEX_UK	IMAG.Glasgow_Norms	PoS.SUBTLEX_UK
sandwich	0.1277847	8	4.246820	6.7647	noun
trousers	0.2015596	8	4.244371	6.6286	noun
wardrobe	0.3255001	8	4.104315	6.6176	noun
diamonds	0.3257108	8	4.063650	6.7813	noun
clothing	0.3302431	8	4.135068	6.5455	noun
calendar	0.3359636	8	4.329810	6.4000	noun
magazine	0.3522379	8	4.287246	6.3846	noun
computer	0.3552034	8	4.641155	6.9259	noun
mountain	0.3577820	8	4.644931	6.8148	noun
bungalow	0.3726770	8	4.172277	6.4242	noun

4.2 Matching by Similarity

You may also want to match by similarity to the target word.

4.2.1 Orthographic similarity

Here’s an example, matching “leaflet” by orthographic similarity (Levenshtein distance). We just have to calculate the similarity measure before using the match_item() function.

library(LexOPS)
library(stringdist)
library(dplyr)

target_word <- "interesting"

suggested_matches <- lexops |>
  mutate(orth_sim = stringdist(string, target_word, method="lv")) |>
  match_item(target = target_word, orth_sim = 0:3)

Some of these entries are misspellings or unusual words, but we could remove these by filtering (e.g. with dplyr::filter()) or matching (with match_item()) by frequency, proportion known, or familiarity ratings.

string	euclidean_distance	orth_sim
interestin	0.8288949	1
interacting	1.6577897	2
intercepting	1.6577897	2
interestingay	1.6577897	2
interestingly	1.6577897	2
interjecting	1.6577897	2
intermeshing	1.6577897	2
intersecting	1.6577897	2
uninteresting	1.6577897	2
entreating	2.4866846	3

4.2.2 Phonological Similarity

To match by phonological similarity, we just have to calculate the Levenshtein distance on one-letter phonemic representations, e.g. with CMU.1letter or eSpeak.br_1letter. Here we find words that are only 0 to 2 phonemic insertions, deletions, or substitutions away from “interesting”.

library(LexOPS)
library(stringdist)
library(dplyr)

target_word <- "interesting"

# get the target word's pronunciation
target_word_pron <- lexops |>
  filter(string == target_word) |>
  pull(CMU.1letter)

# find phonologically similar words
suggested_matches <- lexops |>
  mutate(phon_sim = stringdist(CMU.1letter, target_word_pron, method="lv")) |>
  match_item(target_word, phon_sim = 0:2)

Which gives us:

string	euclidean_distance	phon_sim	CMU.1letter
entrusting	0.9515925	1	EntrAstIG
encrusting	1.9031850	2	EnkrAstIG
entrusted	1.9031850	2	EntrAstId
instructing	1.9031850	2	InstrAktIG
interest	1.9031850	2	IntrAst
interested	1.9031850	2	IntrAstAd
interests	1.9031850	2	IntrAsts
interrupting	1.9031850	2	IntRAptIG
intrastate	1.9031850	2	IntrAstet
mistrusting	1.9031850	2	mIstrAstIG