Navneeth017/malayalam-syntactic-features

Malayalam Syntactic Feature Dataset

A large word-level resource for Malayalam syntax and morphology: 1,901,014 words in 152,454 sentences. Every word has a part-of-speech tag, its lemma (dictionary form), and seven grammatical features: gender, number, tense, aspect, modality, case and polarity.

Malayalam is strongly agglutinative. A single written word such as വാർത്തകളിൽ ("in the news reports") packs together a stem (വാർത്ത), plural number and locative case. This dataset unpacks that information into clean, separate columns. It's ready for training and evaluating part-of-speech taggers, morphological analysers, lemmatisers and feature predictors, and for linguistic study of Malayalam grammar at scale.

Sentences	152,454
Words (tokens)	1,901,014
Unique word forms	278,052
Unique lemmas	40,284
Part-of-speech tags	35
Grammatical features	7

Columns

Column	Description
sent_id	Identifier of the sentence the word belongs to (ml_pos_0000001, …).
tok_id	Position of the word within its sentence, starting at 1. Together with sent_id, it lets you rebuild every sentence in order.
word	The word exactly as it appears in the sentence.
pos_tag	The word's part-of-speech tag: its grammatical category and syntactic role (see Part-of-speech tags).
lemma	The dictionary form of the word, with inflection removed: വാർത്തകളിൽ → വാർത്ത, ആഗ്രഹിച്ച → ആഗ്രഹിക്കുക, അവളുടെ → അവൾ.
gender	Grammatical gender: Masc, Fem or Neut. In Malayalam, gender is expressed in pronouns (അവൻ "he", അവൾ "she", അത് "it"); verbs don't change for gender.
number	Sing (one) or Plur (more than one), e.g. കുട്ടി / കുട്ടികൾ.
tense	Time of the action: Past, Pres (present) or Fut (future).
aspect	How the action unfolds in time: completed, habitual or repeated (values below).
modality	The speaker's stance towards the action: command, purpose, promise, condition, wish, permission… (values below).
case	The grammatical role a noun or pronoun plays in the sentence: subject, object, recipient, owner, location… (values below).
polarity	Neg for negated forms (e.g. -ില്ല "not", -ാതെ "without").

An empty cell means the feature doesn't apply to that word (for example, tense on a noun, or case on a verb).

Example (first five words of the first sentence):

word	pos_tag	lemma	number	tense	modality	case
കന്യാസ്ത്രീയാകാൻ	VERB_INF	ആകുക			purposive
ആഗ്രഹിച്ച	VERB_ADJPTCP	ആഗ്രഹിക്കുക		Past
യുവതി	NOUN_COM	യുവതി	Sing			Nom
വാർത്തകളിൽ	NOUN_COM	വാർത്ത	Plur			Loc
ഇടംപിടിച്ചു	VERB_FIN	പിടിക്കുക		Past

Part-of-speech tags

A part-of-speech (POS) tag says what kind of word something is and what role it can play in a sentence. POS tags are the foundation of syntactic analysis. Parsers, named-entity recognisers, machine translation, grammar checkers and information extraction all rely on them to know which words can be subjects, which mark relations, and where clauses begin and end. The tagset here is designed around Malayalam, giving its distinctive verb forms (participles, conditional forms, the quotative) their own tags.

Tag	Name	Examples	Significance
Nouns
NOUN_COM	Common noun	യുവതി, സംഭവം	Heads most phrases; typical subjects and objects.
NOUN_PROP	Proper noun	പാകിസ്താൻ, ഭാരത	Names of people, places and organisations; central to entity recognition.
NOUN_LOCTIME	Location/time noun	താഴെ, മുന്നിൽ	Nouns that express spatial or temporal relations ("below", "in front").
NOUN_VERBAL	Verbal noun	ഖനനം	An action expressed as a noun ("mining").
Pronouns and determiners
PRON_PERS	Personal pronoun	നാം, അദ്ദേഹം, അത്	Stands in for a noun; key for tracking who is being talked about.
PRON_REFL	Reflexive pronoun	താൻ, സ്വയം	Refers back to the subject ("oneself").
PRON_RECIP	Reciprocal pronoun	പരസ്പരം, തമ്മിൽ	Expresses mutual relations ("each other").
DET_DEM	Demonstrative	ഈ, ആ	Points to a specific thing ("this", "that").
QWORD	Question word	എന്ത്, എങ്ങനെ	Forms questions and indefinite expressions.
Verbs
VERB_FIN	Finite verb	ചെയ്തു, ഇടംപിടിച്ചു	Completes a clause. Malayalam is verb-final, so these mark the end of a sentence or clause.
VERB_ADJPTCP	Adjectival participle	ആഗ്രഹിച്ച, മാറിയ	A verb form that modifies the noun after it; how Malayalam builds relative clauses ("the woman who wished").
VERB_CONJPTCP	Conjunctive participle	കുനിഞ്ഞ്, ഉപയോഗിച്ച്	Chains actions together ("having bent, …", "using …"); links clauses into a sequence.
VERB_INF	Infinitive	മാറാൻ, ഉണ്ടാവുകയും	Purpose and complement clauses ("in order to change").
VERB_COND	Conditional/temporal	ആയാൽ, എത്തുമ്പോൾ	"If" and "when" clauses.
VERB_GER	Gerund	നീന്തൽ	The action itself used as a noun ("swimming").
VERB_NONFIN	Other non-finite verb	ചെയ്യുന്നതിന്റെ	Nominalised verb forms ("of doing").
VERB_NOMPTCP	Nominalised participle	ചെയ്യുന്നതിന്	A participle used as a case-marked noun ("for doing").
Modifiers
ADJ	Adjective	ഫലപ്രദമായ, കഴിഞ്ഞ	Describes a noun.
ADV	Adverb	പിന്നീട്, വീണ്ടും	Describes an action: time, manner, frequency.
INTENS	Intensifier	വളരെ, ഏറ്റവും	Expresses degree ("very", "most").
Numbers and quantity
NUM_CARD	Cardinal number	ഒരു, അഞ്ചു	Counting numbers.
NUM_ORD	Ordinal number	മൂന്നാം, രണ്ടാം	Position in a sequence ("third").
QUANT	Quantifier	ചില, പല, അനേകം	Non-exact amounts ("some", "many").
Particles and connectives
POSTP	Postposition	മുതൽ, ശേഷം	Relation words placed after the noun; Malayalam's equivalent of prepositions.
CONJ_COORD	Coordinating conjunction	പക്ഷേ	Joins clauses of equal status ("but").
CONJ_SUBORD	Subordinating conjunction	എങ്കിലും	Introduces a dependent clause ("although").
QUOT	Quotative marker	എന്ന്, എന്ന	Marks reported speech, thought and naming; extremely common in Malayalam.
NEG	Negation	ഇല്ല, അല്ല	Reverses meaning; essential for sentiment analysis and inference.
EMPH	Emphatic particle	തന്നെ	Adds focus ("itself", "indeed").
ENUM	Enumerative clitic	എന്നിവർ, എന്നീ	Closes a list ("A, B and C").
INTJ	Interjection	അയ്യോ, എടോ	Exclamations and calls.
Symbols and other
SYM	Symbol	. - ,	Punctuation and symbols.
PUNCT	Punctuation	.	Punctuation marks.
FOREIGN	Foreign word	—	Words from other languages.
RESID	Residual	—	Anything not covered by the categories above.

Grammatical features

Malayalam marks who did what to whom, when and how with suffixes, not word order. A noun's case suffix tells you whether it's the subject, object or location. A verb's suffixes tell you the time of the action, whether it's complete, and the speaker's attitude. Having these as explicit features lets models and rule systems use them directly, which is especially valuable for a morphologically rich, low-resource language.

Feature	Values (word counts)	Significance
case	Nom 532,003 · Loc 110,419 · Dat 57,672 · Gen 52,536 · Acc 33,417 · Soc 9,094 · Voc 4,016 · Ins 1,275 · Abl 754	The role of a noun in the sentence: Nom: subject Acc (-നെ): direct object Dat (-ക്ക്, -ന്): recipient Gen (-ന്റെ, -ഉടെ): owner Loc (-ൽ): place Soc (-ഓട്): companion / "with" Ins: instrument / "by means of" Abl: source / "from" Voc: addressing someone directly
number	Sing 657,591 · Plur 101,667	One vs many (plural -കൾ, -മാർ). Needed for agreement and counting.
tense	Past 153,312 · Pres 49,173 · Fut 25,639	When the action takes place. Key for event ordering and timelines.
aspect	simple-perfect 3,179 · remote-perfect 2,678 · cont-perfect 1,161 · habitual 762 · iterative 124 · iterative-past 119 · iterative-present 188 · iterative-future 13 · emphatic-iterative 20	How the action unfolds: perfect: completed, with present relevance (-ിരുന്നു, -ിട്ടുണ്ട്) habitual: regularly done (-ാറുണ്ട്) iterative: repeated or ongoing (-കൊണ്ടിരിക്കുന്നു)
modality	purposive 22,092 · imperative 7,755 · promissive 4,393 · conditional 2,347 · permissive 969 · optative 312 · compulsive 89 · desiderative 57 · satisfactive 37 · quotative 15 · irrealis 14 · precative 11 · precative-neg 5 · abilitative 1	The speaker's stance towards the action: purposive: purpose (-ആൻ, "in order to") imperative: command or need (-ണം) promissive: offer or promise (-ാം) conditional: condition (-ആൽ, "if") permissive: permission optative: wish (-ട്ടെ, "let it") compulsive: obligation desiderative: desire irrealis: hypothetical ("would have") precative: polite request
gender	Neut 8,640 · Masc 8,263 · Fem 600	Masculine / feminine / neuter, expressed in pronouns. Useful for coreference: knowing who "he", "she" or "it" refers to.
polarity	Neg 10,598	Marks negated forms. Important for meaning, sentiment and question answering.

Usage

from datasets import load_dataset

ds = load_dataset("Navneeth017/malayalam-syntactic-features", split="train")
print(ds[0])
# {'sent_id': 'ml_pos_0000001', 'tok_id': 1, 'word': 'കന്യാസ്ത്രീയാകാൻ', 'pos_tag': 'VERB_INF',
#  'lemma': 'ആകുക', 'gender': '', 'number': '', 'tense': '', 'aspect': '', 'modality': 'purposive',
#  'case': '', 'polarity': ''}

Rebuilding sentences (e.g. for sequence labelling):

import pandas as pd
df = ds.to_pandas()
sentences = df.sort_values(["sent_id", "tok_id"]).groupby("sent_id").agg(list)
# sentences.loc["ml_pos_0000001", "word"], sentences.loc["ml_pos_0000001", "pos_tag"], ...

Using the TSV: load it with quote handling turned off, since some tokens are a bare " character:

df = pd.read_csv("malayalam_syntactic_features.tsv", sep="\t", quoting=3, dtype=str, keep_default_na=False)

Suggested tasks

• Part-of-speech tagging: predict pos_tag for each word.
• Lemmatisation: predict lemma from word.
• Morphological feature prediction: predict case, number, tense, aspect, modality, gender and polarity, as one multi-task model or one model per feature.
• Morphologically aware pre-training and probing: test whether a language model's representations encode Malayalam case, tense and mood.
• Linguistic analysis: frequency and distribution of grammatical categories in Malayalam text.

Files

File	Description
data/train.parquet	The full dataset in Parquet format (used by the dataset viewer and load_dataset).
malayalam_syntactic_features.tsv	The same data as a tab-separated text file.

License

This dataset is released under the Creative Commons Attribution 4.0 International (CC BY 4.0) license. You are free to use, share and adapt it, including for commercial purposes, as long as you give appropriate credit.