It was recently found that letter-shapes have a non-arbitrary relation with their canonical pronunciations, in multiple orthographies and quantified across the whole of each orthography: letters that look similar tend to have similar pronunciations. Similarly, there is phonosemantic systematicity at the word level: words that sound similar tend to have similar meanings. We investigated for the first time whether a similar systematicity exists in Chinese characters. We measured all the pairwise phonological distances and all the pairwise orthographic distances of the 58 Chinese pictographic characters that are taught to year 1 and 2 in Chinese primary schools. The correlation was tested between the two lists of distances and verified by a Mantel test. We found a significant negative correlation between characters and their segmental pronunciations: characters that look similar tend to have dissimilar segmental pronunciations. This contrasts with the positive correlations found in previous similar research with alphabetic writing. We conclude, first, that questions of systematicity in the Chinese writing system are tractable in the same terms and by the same methodology as that applied to alphabetic writing systems. Second, segment-based processing requires to be augmented by tones for there to be systematicity that is comparable to that found in alphabetic writing systems. Any non-arbitrary relation between letter shapes and sounds may help bootstrap the acquisition of literacy.

Grapho-Syllabic Systematicity in Chinese: Chinese Pictographs Have a Non-Arbitrary relation with their Pronunciations

A challenge in learning phonological grammars is learning how phonological processes interact. It has been argued that some process interactions are easier to learn than others. One basis for this argument is asymmetries observed in experimental settings: artificial languages generated from certain process interactions are more likely to be successfully reproduced by participants than others. In this paper, we argue that asymmetries in production do not necessarily provide direct support that some phonological interactions are easier to learn. Rather, we show that these asymmetries can instead emerge due to differences in the number of consistent or nearly-consistent grammars each pattern has. We present a noisy channel model of morpho-phonological learning and apply it to a recent behavioral study examining the learnability of phonological process interactions. We find that, due to the relative difference in the number of grammars that can exactly match or nearly match the observed data, the model achieves the same qualitative results as those observed in experimental settings.

Assessing the learnability of process interactions using grammatical spaces

We present proof-of-concept modeling results which demonstrate a novel way to evaluate theories of the structure of the phonological grammar: We generate predictions of how the behavior of constraint-based grammars and analogical models is likely to change under processing load. Previous results (Moore-Cantwell and Kush, 2019) show systematic differences in participant behavior when generating novel forms under working memory load, vs not. Under load, participants observed fewer generalizations of the language, produced less intra-speaker variation, and differed more participant-to-participant. We modeled memory load in constraint-based grammar (MaxEnt, Goldwater and Johnson, 2003) via probabilistic ignoring of constraints. Memory load in analogy (TiMBL, Daelemans et al., 2018) was modeled as a restricting of the search space for an analogical set, based on word-frequency. Our constraint-based model predicted a more systematic breakdown, more in line with previous experimental findings, while our analogical model predicted more word-to-word variation.

Modeling grammatical breakdown under processing load

First language learners use various linguistic cues to construct their language(s)’ grammar. I study how phonological and morphological structures interact in the minds of two young learners of Russian (1;6 - 2;11 years of age) and their caretakers. While phonology is traditionally recognized to develop ahead of morphology, my data show a ‘backward’ directionality for the younger child only: the grasp of consonant cluster inventory depends on the rising morphological complexity, measured as a number of morphemes per word.
This interaction is first backed by consonant clustering, being a natural result of phonological alternations at the morpheme boundary; and second, by language pathology data, an aphasic patient attempting to improve the coda sonority in inflected words but not monomorphemic ones during speech production.
If this result is validated on a larger sample, the finding is meaningful both theoretically and practically for the speech therapy procedures.

Morphological Strengthening of Phonological Performance in Child Speech

In conversation, interlocutors rapidly exchange well-timed turns-at-talk, generally starting to plan their contributions already during the incoming turns. Previous research showed that while this early-planning-strategy supports fast turn-taking, it comes with increased planning difficulty compared to planning after the incoming turn. The present study investigates whether concurrent planning is also detrimental to speech input comprehension, targeting the level of phonological processing. 60 subjects conducted a dual-task experiment, verbally responding to auditorily presented quiz-questions while simultaneously monitoring them for pre-specified target phonemes (60% present), pressing the space-bar upon encounter. Subjects’ monitoring performance for phonemes encountered during response planning was significantly worse than for phonemes encountered when response planning was not yet possible. Additionally, phoneme detection latencies were significantly longer when concurrently planning a verbal response than in a single-task phoneme-detection control experiment. Thus, results indicate that parallel speech planning, while common, leads to decreased input processing, possibly due to interference effects.

Reduced phonological processing during speech-planning

Arabic is a morphologically rich and complex language, with numerous dialectal variants. Previous efforts on Arabic morphology modeling focused on specific variants and specific domains using a range of techniques with different degrees of linguistic modeling transparency. 
In this paper we propose a new approach to modeling Arabic morphology with an eye towards multi-dialectness, resource openness, and easy extensibility and use.
We demonstrate our approach by modeling verbs from Standard Arabic and Egyptian Arabic, within a common framework, and with high coverage.

Morphotactic Modeling in an Open-source Multi-dialectal Arabic Morphological Analyzer and Generator

Recent advances in the field of deep learning have led to a growing interest in the development of NLP approaches for low-resource and endangered languages. Nevertheless, relatively little research, related to NLP, has been conducted on indigenous languages. These languages are considered to be filled with complexities and challenges that make their study incredibly difficult in the NLP and AI fields. This paper focuses on the morphological segmentation of indigenous languages, an extremely challenging task because of polysynthesis, dialectal variations with rich morpho-phonemics, misspellings and resource-limited scenario issues. The proposed approach, towards a morphological segmentation of Innu-Aimun, an extremely low-resource indigenous language of Canada, is based on deep learning. 
Experiments and evaluations have shown promising results, compared to state-of-the-art rule-based and unsupervised approaches.

Deep Learning-Based Morphological Segmentation for Indigenous Languages: A Study Case on Innu-Aimun

Word embeddings are growing to be a crucial resource in the field of NLP for any language. This work introduces a novel technique for static subword embeddings transfer for Indic languages from a relatively higher resource language to a genealogically related low resource language. We primarily work with Hindi-Marathi, simulating a low-resource scenario for Marathi, and confirm observed trends on Nepali. We demonstrate the consistent benefits of unsupervised morphemic segmentation on both source and target sides over the treatment performed by fastText. Our best-performing approach uses an EM-style approach to learning bilingual subword embeddings; we also show, for the first time, that a trivial ``copy-and-paste'' embeddings transfer based on even perfect bilingual lexicons is inadequate in capturing language-specific relationships. We find that our approach substantially outperforms the fastText baselines for both Marathi and Nepali on the Word Similarity task as well as WordNet-Based Synonymy Tests; on the former task, its performance for Marathi is close to that of pretrained fastText embeddings that use three orders of magnitude more Marathi data.

Subword-based Cross-lingual Transfer of Embeddings from Hindi to Marathi and Nepali

OSU’s inflection system is a transformer whose
input is augmented with an analogical exemplar
showing how to inflect a different word into the
target cell. In addition, alignment-based heuris-
tic features indicate how well the exemplar is
likely to match the output. OSU’s scores sub-
stantially improve over the baseline transformer
for instances where an exemplar is available,
though not quite matching the challenge win-
ner. In Part 2, the system shows a tendency to
over-apply the majority pattern in English, but
not Arabic.

OSU At at SIGMORPHON ‘22: Analogical Inflection with Rule Features

Though recently there have been an increased interest in how pre-trained language models encode different linguistic features, there is still a lack of systematic comparison between languages with different morphology and syntax. In this paper, using BERT as an example of a pre-trained model, we compare how three typologically different languages (English, Korean, and Russian) encode morphology and syntax features across different layers. In particular, we contrast languages which differ in a particular aspect, such as flexibility of word order, head directionality, morphological type, presence of grammatical gender, and morphological richness, across four different tasks.

Cross-linguistic comparison of linguistic feature encoding in BERT models
for typologically different languages

This work describes an implementation of the “extended alignment” model for cognate reflex prediction submitted to the “SIGTYP 2022 Shared Task on the Prediction of Cognate Reflexes”. Similarly to List et al. (2022a), the technique involves an automatic extension of sequence alignments with multilayered vectors that encode informational tiers on both site-specific traits, such as sound classes and distinctive features, as well as contextual and suprasegmental ones, conveyed by cross-site referrals and replication. The method allows to generalize the problem of cognate reflex prediction as a classification problem, with models trained using a parallel corpus of cognate sets. A model using random forests is trained and evaluated on the shared task for reflex prediction, and the experimental results are presented and discussed along with some differences to other implementations.

Reflex Prediction Through Extended Alignments

We present NEAMER - Named Entity Augmented Multi-word Expression Recognizer. This system is inspired by non-compositionality characteristics shared between Named Entity and Idiomatic Expressions. We utilize transfer learning and locality features to enhance idiom classification task. This system is our submission for SemEval Task 2: Multilingual Idiomaticity Detection and Sentence Embedding Subtask A OneShot shared task. We achieve SOTA with F1 0.9395 during post-evaluation phase. We also observe improvement in training stability. Lastly, we experiment with non-compositionality knowledge transfer, cross-lingual fine-tuning and locality features, which we also introduce in this paper.

Underline digital video library

Grapho-Syllabic Systematicity in Chinese: Chinese Pictographs Have a Non-Arbitrary relation with their Pronunciations

Assessing the learnability of process interactions using grammatical spaces

Modeling grammatical breakdown under processing load

Morphological Strengthening of Phonological Performance in Child Speech

Reduced phonological processing during speech-planning

Morphotactic Modeling in an Open-source Multi-dialectal Arabic Morphological Analyzer and Generator

Deep Learning-Based Morphological Segmentation for Indigenous Languages: A Study Case on Innu-Aimun

Subword-based Cross-lingual Transfer of Embeddings from Hindi to Marathi and Nepali

OSU At at SIGMORPHON ‘22: Analogical Inflection with Rule Features

Cross-linguistic comparison of linguistic feature encoding in BERT models for typologically different languages

Reflex Prediction Through Extended Alignments

NEAMER - Named Entity Augmented Multi-word Expression Recognizer

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES