In derivational morphology, what mechanisms govern the variation in form-meaning relations between words? The answers to this type of questions are typically based on intuition and on observations drawn from limited data, even when a wide range of languages is considered. Many of these studies are difficult to replicate and generalize. To address this issue, we present GLeMM, a new derivational resource designed for experimentation and data-driven description in morphology. GLeMM is characterized by (i) its large size, (ii) its extensive coverage (currently amounting to seven European languages, i.e., German, English, Spanish, French, Italian, Polish, Russian, (iii) its fully automated design, identical across all languages, (iv) the automatic annotation of morphological features on each entry, as well as (v) the encoding of semantic descriptions for a significant subset of these entries. It enables researchers to address difficult questions, such as the role of form and meaning in word-formation, and to develop and experimentally test computational methods that identify the structures of derivational morphology. The article describes how GLeMM is created using Wiktionary articles and presents various case studies illustrating possible applications of the resource.
Figures & tables
GLAWI
Kaikki
MorphyNet
definitions
morphological
definitions
morphological
entries
section
section
English
yes
yes
yes
yes
yes
French
yes
yes
yes
yes
yes
Italian
yes
yes
yes
German
yes
yes
yes
Table 1: Source of the candidate pairs used to create the GLeMM lexicons
A
B
C
D
left pattern
right pattern
positivist
positivism
feminist
feminism
ˆ (.+)ist$
ˆ (.+)ism$
positivist
positivism
feminist
feminism
ˆ (.+)i(.+)ist$
ˆ (.+)i(.+)ism$
positivist
positivism
activist
activism
ˆ (.+)itivist$
ˆ (.+)tivism$
positivist
positivism
activist
activism
ˆ (.+)tivist$
ˆ (.+)tivism$
positivist
positivism
activist
activism
ˆ (.+)ivist$
ˆ (.+)ivism$
positivist
positivism
activist
activism
ˆ (.+)vist$
ˆ (.+)vism$
Table 2: Examples of alternation patterns yielded by analogies involving the pair positivist N and positivism N
lemma1
cat1
lemma2
cat2
stem
exponent1
exponent2
rastrero
A
rastreramente
R
rastrer
ˆ (.+)o$
ˆ (.+)amente$
‘despicable.M’
‘despicably’
ducha
N
duchita
N
duch
ˆ (.+)a$
ˆ (.+)ita$
‘shower’
‘small shower’
talentoso
A
talento
N
talento
ˆ (.+)so$
ˆ (.+)$
‘talented.M’
‘talent’
Table 3: Examples of entries in the Spanish pairs.tsv table
lemma1
cat1
lemma2
cat2
stem
exponent1
exponent2
sensitivity
N
electrosensitivity
N
sensitivity
ˆ (.+)$
ˆ electro(.+)$
clonally
R
subclonally
R
clonally
ˆ (.+)$
ˆ sub(.+)$
supertechnological
A
technology
N
technolog
ˆ super(.+)ical$
ˆ (.+)y$
Table 4: Examples of entries in the English pairs.tsv table
a.
spry A spryness N spry ˆ (.+)ˆ(.+)ness
definition2
The property of being spry.
lemmatized_definition2
the property of be spry .
b.
astounding A unastounding A astounding ˆ (.+)ˆun(.+)
definition2
Not astounding.
lemmatized_definition2
not astounding .
c.
concretize V concretization N concretiz ˆ (.+)eˆ(.+)ation
Table 5: Examples of entries in the English defs.tsv table
a.
coffrage N décoffrer V coffr ˆ (.+)ageˆdeˊ(.+)er
‘formwork’ ‘unmould’
definition2
Retirer le coffrage après prise du béton.
‘Remove the formwork after the concrete has set.’
lemmatized_definition2
retirer le coffrage après prise de le béton .
b.
neutralité N neutraliser V neutral ˆ (.+)itéˆ(.+)iser
‘neutrality’ ‘neutralize’
Table 6: Examples of entries in the French defs.tsv table
ordered pairs
defs
ratio
pair sets
defs
ratio
English
779,170
271,496
0.348
395,886
254,528
0.643
Russian
732,633
261,713
0.357
366,352
250,425
0.684
French
271,343
86,526
0.319
135,679
76,686
0.565
Polish
196,433
35,581
0.181
98,222
31,533
0.321
German
136,248
42,438
0.311
68,127
39,995
0.587
Italian
121,400
9,569
0.079
60,700
8,843
0.146
Table 7: Number of pairs of lexemes, with and without semantic descriptions, in the GLeMM lexicons
correct
optimal
pairs
accuracy
patterns
accuracy
English
94
0.94
87
0.93
French
99
0.99
92
0.93
Italian
100
1.00
84
0.84
Table 8: Accuracy of the pairs and optimality of the exponents
resources
languages
objects
CELEX
German, English, Dutch
complex lexemes
Démonette
French
relations (set-pairs)
Derivatario
Italian
complex lexemes
Derinet
Czech, Russian, Polish, Spanish
complex lexemes
DerivBase
German, Russian
first-level relations
MorphyNet
15 languages
relations (set-pairs)
Table 9: Objects of the existing resources considered for the quantitative comparison of GLeMM lexicons
Language
GLeMM
UDer
MorphyNet
UniMorph
CELEX
Other
English
395,886
32,483
225,131
225,131
52,447
Russian
366,352
303,615
93,039
14,048
French
135,679
13,808
72,952
73,259
111,059
(F)
Polish
98,222
214,093
58,711
58,742
German
68,127
43,367
29,381
29,381
51,728
63,479
(G)
Italian
60,700
1,783
58,848
11,147
(I)
Table 10: Size comparison between GLeMM and other resources
lexeme1
lexeme2
derivational type
MorphyNet
GLeMM
manage V
manager N
direct derivation
yes
yes
measurable ADJ
measurability N
direct derivation
yes
yes
characterize V
characterizability N
multiple derivations
yes
yes
hospitalize V
rehospitalization N
multiple derivations
no
yes
minimalist N
minimalism N
indirect (cross) relation
no
yes
productive ADJ
productible ADJ
indirect (cross) relation
no
yes
Table 11: Comparison of GLeMM ( pairs.tsv ) and MorphyNet for a few pairs of lexemes whose relations illustrate various derivational configurations.
mutually motivated pairs
defined pairs
ratio
English
4,368
256,712
0.017
French
2,422
77,897
0.031
German
650
40,320
0.016
Italian
314
9,000
0.035
Polish
1,140
32,103
0.036
Russian
3,178
252,014
0.013
Table 12: Number of mutually motivated pairs in GLeMM lexicons
transitive pairs
two-edge paths
ratio
English
10,573
179,290
0.059
French
4,557
40,288
0.113
German
404
17,305
0.023
Italian
175
1,987
0.088
Polish
857
11,977
0.072
Russian
8,508
206,200
0.041
Table 13: Number of transitive relations in the GLeMM lexicons
cycles
two-edge paths
ratio
English
702
179,290
0.004
French
390
40,288
0.010
German
24
17,305
0.001
Italian
30
1,987
0.015
Polish
141
11,977
0.012
Russian
762
206,200
0.004
Table 14: Number of 3-edge cycles in GLeMM lexicons
We present MorfFlex, a morphological dictionary architecture suitable for languages with extensive regularity in both inflection and derivation. As the primary example of MorfFlex in use we introduce MorfFlex CZ, a morphological dictionary of Czech. It is distributed as a simple, unstructured list of <wordform, lemma, tag> triplets, however, its manually maintained, unpublished source files and conversion scripts encode a sophisticated system of inflectional and derivational patterns. These patterns dramatically reduce the otherwise enormous size of the dictionary, which currently contains over 100 million wordforms and more than 1 million lemmas. The MorfFlex CZ dictionary serves as an essential resource for ensuring the consistency of manual morphological annotation in the Prague Dependency Treebanks and underpins state-of-the-art automatic tools such as MorphoDiTa. In this paper, we focus on: (i) presenting an effective method for managing the rich morphological system within the dictionary, and (ii) demonstrating the utility of such a language resource for maintaining annotation consistency in corpora and supporting the development of advanced NLP applications.
Jaroslava Hlaváčová, Marie Mikulová, Barbora Štěpánková +2
Charles University, Faculty of Mathematics and Physics, Institute of Formal and Applied Linguistics, Malostranské náměstí 25, 118 00 Prague 1, Czech Republic
Arabic morphology remains challenging for large language models, since fluent generation does not guarantee accurate morphosyntactic control. Existing Arabic evaluations mainly target downstream tasks and do not directly test controlled morphological generation from explicit lexical and feature-based input. We introduce YallaMorph, a large-scale benchmark for Arabic morphological generation covering verbs, nouns, adjectives, their cliticized forms, and invalid configurations. We evaluate multilingual and Arabic-oriented LLMs under diacritized and undiacritized settings over 600K benchmark entries. Results show that Arabic morphological generation remains difficult, especially for cliticized, unseen, and morphologically rare forms.
Mahmoud Reda, Salam Khalifa, Reham Marzouk +1
New York University Abu Dhabi · Stony Brook University · Mohamed bin Zayed University of Artificial Intelligence
While multilingual large language models (LLMs) perform well on high-level tasks like translation and question answering, their ability to handle grammatical gender and morphological agreement remains underexplored. In morphologically rich languages, gender influences verb conjugation, pronouns, and even first-person constructions with explicit and implicit mentions of gender. We introduce MORPHOGEN, a morphologically grounded large-scale benchmark dataset for evaluating gender-aware generation in three typologically diverse grammatically gendered languages: French, Arabic, and Hindi. The core task, GENFORM, requires models to rewrite a first-person sentence in the opposite gender while preserving its meaning and structure. We construct a high-quality synthetic dataset spanning these three languages and benchmark 15 popular multilingual LLMs (2B-70B) on their ability to perform this transformation. Our results reveal significant gaps and interesting insights into how current models handle morphological gender. MORPHOGEN provides a focused diagnostic lens for gender-aware language modeling and lays the groundwork for future research on inclusive and morphology-sensitive NLP.
Mehul Agarwal, Aditya Aggarwal, Arnav Goel +2
SBILab, Indraprastha Institute of Information Technology Delhi