[Paper Review] Comparing morphological complexity of Spanish, Otomi and Nahuatl

Ximena Gutierrez-Vasques, Víctor Mijangos|arXiv (Cornell University)|Aug 13, 2018

Natural Language Processing Techniques13 references3 citations

TL;DR

This study compares morphological complexity in Spanish, Otomi, and Nahuatl using corpus-based metrics: type-token ratio (TTR), entropy, and perplexity across two small parallel corpora. It reveals that while Nahuatl exhibits high morphological productivity (high TTR), its morphological sequences are more predictable (lower entropy/perplexity) than Spanish; Otomi, however, shows high complexity in both productivity and unpredictability, indicating greater morphological intricacy due to complex inflectional classes, stem alternations, and derivational processes.

ABSTRACT

We use two small parallel corpora for comparing the morphological complexity of Spanish, Otomi and Nahuatl. These are languages that belong to different linguistic families, the latter are low-resourced. We take into account two quantitative criteria, on one hand the distribution of types over tokens in a corpus, on the other, perplexity and entropy as indicators of word structure predictability. We show that a language can be complex in terms of how many different morphological word forms can produce, however, it may be less complex in terms of predictability of its internal structure of words.

Motivation & Objective

• To quantify morphological complexity in three typologically distinct languages—Spanish (Indo-European), Otomi (Otomian), and Nahuatl (Uto-Aztecan)—using corpus-based metrics.
• To investigate whether high morphological productivity (many word forms) correlates with high predictability of morphological sequences.
• To evaluate the impact of morphological normalization (lemmatization, stemming, segmentation) on complexity measures.
• To explore whether low-resource languages like Otomi and Nahuatl exhibit different complexity profiles than high-resource languages like Spanish.
• To assess the complementarity of TTR (productivity) and entropy/perplexity (predictability) as quantitative measures of morphological complexity.

Proposed method

• The study uses two small parallel corpora in Spanish, Otomi, and Nahuatl to ensure comparable semantic content across languages.
• Type-token ratio (TTR = types/tokens) is computed to measure morphological productivity and word-form diversity.
• Statistical language models are trained on sequences of morphs to compute entropy and perplexity as measures of predictability in morphological structures.
• Different morphological normalization techniques (lemmatization, stemming, segmentation) are applied to assess their impact on TTR and predictability metrics.
• The analysis compares results across languages to evaluate differences in morphological complexity dimensions.
• Entropy and perplexity are calculated over morpheme sequences to assess uncertainty and predictability in word formation.

Experimental results

Research questions

• RQ1 How does the type-token ratio (TTR) of Spanish, Otomi, and Nahuatl compare in parallel corpora, indicating morphological productivity?
• RQ2 To what extent is the sequence of morphs in each language predictable, as measured by entropy and perplexity?
• RQ3 How do different morphological normalization techniques affect the measured complexity of each language?
• RQ4 Does a language with high morphological productivity (high TTR) necessarily exhibit low predictability (high entropy)?
• RQ5 How do the morphological complexity profiles of Otomi and Nahuatl differ from that of Spanish, especially considering their linguistic families and resource levels?

Key findings

• Nahuatl exhibits the highest TTR among the three languages, indicating high morphological productivity due to its agglutinative and polysynthetic nature.
• Despite high TTR, Nahuatl shows lower entropy and perplexity than Spanish, suggesting its morphological sequences are more predictable.
• Otomi displays high TTR and also high entropy and perplexity, indicating high complexity in both productivity and unpredictability of morphological structures.
• The high entropy in Otomi is attributed to complex inflectional classes, stem alternations, prefix changes, and tone-sensitive morphology, especially in verbs.
• Derivational processes in Otomi contribute to unpredictability, as they are less frequent and less regular than inflectional processes.
• The study demonstrates that TTR and predictability (entropy/perplexity) are complementary measures: a language can be highly productive but predictable (Nahuatl), or highly productive and unpredictable (Otomi).