morphology

这篇论文通过在Tagalog上的受控实验（MorphAustro-Tok与BPE及打乱形态素的对照组比较），发现形态感知分词确实能提升机器翻译效果，但形态正确性本身并不贡献收益——增益仅来自分词粒度与一致性，而非语言学上正确的切分。

该论文利用 Pythia 和 OLMo 等开源模型的中间检查点，系统研究了大语言模型在训练过程中形态结构、嵌入漂移、重复主导与共现效应的演化规律，发现模型在语义嵌入稳定之前就已习得形态结构，且训练数据规模有助于发散识别，而共现和重复则有负面影响。

This paper presents the first neural benchmark for Sinhala Sandhi splitting using character-level seq2seq models on the SandhiLex dataset, with a bidirectional LSTM encoder reaching 68.40% exact-match accuracy on hard lexicalized/derivational Sandhi versus 94% on regular affixational cases.

Morphometric Immitation presents a three-stage framework that transforms reconstructed human hand-object interactions into zero-shot sim-to-real visuomotor policies, enhancing contact preservation and success rates in dexterous manipulation tasks.

The paper diagnoses systematic errors in neural morphological generation for Japanese past-tense verb inflection, focusing on a rare irregular subtype that causes disproportionate errors. Ablation experiments demonstrate that removing this subtype improves accuracy more than removing all irregular verbs, emphasizing the importance of fine-grained subclass analysis.

This paper probes character-level transformers to investigate whether they encode the Spanish L-shaped morphome, an irregular morphological pattern, as an abstract class or just surface alternations. The authors find that the encoding is item-specific and localized, but does not generalize like human learners.

Introduces MORFES, a benchmark of 500 expert-verified items for testing productive inflectional competence in Modern Greek, and evaluates open language models including their own Sophea-Genesis-1, which leads on inflectional morphology.

This paper presents Morpheus, a neural tokenizer and word embedder for Turkish that learns morpheme boundaries without string normalization, achieving lossless tokenization and competitive embeddings for lexical retrieval, while using less GPU memory than subword tokenizers.

Introduces PACUTE, a diagnostic benchmark of 4,600 tasks evaluating morphological understanding in Filipino, revealing that even frontier models struggle with morpheme decomposition and productive morphological composition.

A personal account of finding a fossil seashell in the Saudi desert and using computational methods to analyze its shape compared to thousands of shell species.

Presents a novel pattern-and-root model for describing Arabic noun inflection, focusing on broken plurals, with a taxonomy of 160 classes and an encoding scheme applied to 3,200 entries, aiming to improve computational language resources.

This paper investigates how character-level transformer models generalize to irregular verb subtypes in Japanese past-tense inflection. Controlled experiments show that including irregular examples can improve generalization, challenging the assumption that regularity simplifies learning.