A Computational Analysis of Grammatical Change in Classical and Modern Korean Literature

Overview

This study asks whether data science can measure how Korean grammar has changed, comparing classical and modern literary texts quantitatively instead of relying on qualitative reading of selected passages. It began in Grade 10 and was developed into a full paper in Grade 12.

Research Questions

• How have the frequency and function of particles changed between classical and modern Korean literature?
• Are changes in verb conjugation and verbal endings associated with historical shifts?
• Is sentence structure and length related to social change?

Corpus

• Classical: Chunhyangjeon and Heungbujeon, two Joseon-dynasty pansori-based novels, 1,105 sentences in total
• Modern: Cho Se-hui's The Dwarf (166 sentences) and Lee Tae-jun's Moonlit Night (120 sentences), 286 sentences in total
• Sources: the Korea Copyright Commission and AI-Hub

Sentences in the corpus

• Modern
• Classical

• Chunhyangjeon + Heungbujeon1,105
• The Dwarf166
• Moonlit Night120

View data

Item	Value
Chunhyangjeon + Heungbujeon	1,105
The Dwarf	166
Moonlit Night	120

Preprocessing

• Removed special characters, extra whitespace, and sentences containing Chinese characters (hanja)
• Converted archaic expressions to modern Korean orthography
• Segmented every text into sentences

From text to grammatical counts

1. 1

   Collect

   Four works from the Korea Copyright Commission and AI-Hub

2. 2

   Clean

   Remove symbols and hanja sentences, modernize spelling, split sentences

3. 3

   Tag

   Segment morphemes and assign POS tags with Kkma (KoNLPy)

4. 4

   Count

   Tally particles and past-tense, honorific, and connective endings

5. 5

   Compare

   Separate tags shared by both periods from those unique to each work

Method

• Morpheme segmentation and part-of-speech tagging with the Kkma analyzer in KoNLPy (Python)
• Frequency counts for topic particles (eun/neun), subject particles (i/ga), and object particles (eul/reul), along with past-tense, honorific, and connective endings
• A custom algorithm separating the POS tags shared by both corpora from those found only in each work
• Comparison along four dimensions: particle usage, verb conjugation, sentence structure, and diachronic trends shown in graphs

Results

The three most frequent POS tags in each work are listed below, with raw counts.

• Chunhyangjeon: connective ending -go (753), subject particle -i (682), adnominal ending -n (679)
• Heungbujeon: connective ending -go (267), object particle -eul (252), subject particle -i (193)
• The Dwarf: past-tense ending -eot (47), sentence-final ending -da (40), object particle -eul (34)
• Moonlit Night: connective ending -go (29), verb ha- (25), adnominal ending -n (20)
• Words found only in the classical works reflected the Joseon setting, such as jeomgo (roll call) and sikja (literate person), while the modern works used concrete, everyday words such as Yeong-ho (a name), sugeon (towel), and sansam (wild ginseng)

Three most frequent POS tags in each work

• Modern
• Classical

Chunhyangjeon

• -go · connective ending753
• -i · subject particle682
• -n · adnominal ending679

Heungbujeon

• -go · connective ending267
• -eul · object particle252
• -i · subject particle193

The Dwarf

• -eot · past-tense ending47
• -da · sentence-final ending40
• -eul · object particle34

Moonlit Night

• -go · connective ending29
• ha- · verb25
• -n · adnominal ending20

View data

Work	Tag	Count
Chunhyangjeon	-go · connective ending	753
Chunhyangjeon	-i · subject particle	682
Chunhyangjeon	-n · adnominal ending	679
Heungbujeon	-go · connective ending	267
Heungbujeon	-eul · object particle	252
Heungbujeon	-i · subject particle	193
The Dwarf	-eot · past-tense ending	47
The Dwarf	-da · sentence-final ending	40
The Dwarf	-eul · object particle	34
Moonlit Night	-go · connective ending	29
Moonlit Night	ha- · verb	25
Moonlit Night	-n · adnominal ending	20

Findings

• Classical texts linked clauses mainly through connective endings (-go, -n), producing long, complex sentences
• Modern texts relied more on case particles (i/ga, eul/reul) and the past-tense ending -eot, favoring shorter, self-contained sentences
• Korean sentence construction shifted from ending-based clause linkage to compact, particle-based structures, alongside a move toward realistic, dialogue-centered narration

Limitations

• The classical corpus is nearly four times larger than the modern one, so raw counts across corpora should be read with care; normalized frequencies would strengthen the comparison
• Kkma was built for contemporary Korean and may be less accurate on classical texts
• Four works limit how far the findings generalize