Syllabification describes the task of dividing words into syllables. Due to many rules and exceptions, training an algorithm to perform syllabification with high accuracy remains a challenge. Throughout the last decades, different algorithms have been put forth for Dutch syllabification, yet a comprehensive comparative assessment has not been done. Additionally, deep learning has gained significant popularity within NLP in recent years, yet no modern deep-learning based framework has been developed for Dutch orthographic syllabification. Finally, phonetic and orthographic syllabification algorithms have been examined separately, but not in combination. The aim of the current research was twofold: (a) to examine the performance of existing Dutch syllabification algorithms, and (b) to investigate whether combining phonetic and orthographic information into a single model can increase syllabification performance. To compare the performance of algorithms, four algorithms (Brandt Corstius, Liang, Trogkanis-Elkan (CRF), and a newly conceived deep-learning model) were applied to three different datasets (dictionary words, loanwords, pseudowords). The algorithms show varying performance across datasets, with the data-driven algorithms outperforming a knowledge-based algorithm in all but one condition. The new deep-learning methods developed led to increased performance compared to the best found in the literature (99.65% word accuracy, a 0.14% improvement). An analysis of the words for which adding phonetic information improved syllabification performance indicates that these were words in which the orthographic ambiguity could be resolved by information on pronunciation. Future research could examine other areas where phonetic information can benefit orthographic processing. In addition, the newly developed deep learning frameworks can be applied to other languages than Dutch.
Figures & tables
Full Dutch Word
Hyphenated 2 2 2 Retrieved from https://github.com/jitsi/jiwer
Syllabified
Translation
eland
eland
e-land
moose
atoomenergie
atoom-ener-gie
a-toom-e-ner-gie
atomic energy
gloria
glo-ria
glo-ri-a
gloria
aliënatie
ali-ena-tie
a-li-e-na-tie
alienation
bakoven
bak-oven
bak-o-ven
baking oven
bioscoop
bio-scoop
bi-o-scoop
cinema
Table 1: Examples of Dutch words with different syllabification and hyphenation word divisions.
Dataset
Size (n)
Origin
Description
Training
Testing
Dictionary
264,343
29,371
CELEX
Common words found in the Dutch dictionary. The CELEX dataset was used as a training and testing set with a 90%/10% split.
Loanwords
-
1,135
De Sijs (2009)
Words found in the Dutch language originating from other languages, e.g., French or English. Used as a testing set only.
Pseudowords
-
99
CHOREC
Words artificially created to imitate Dutch word conventions. Used as a testing set only.
Table 2: Datasets used for algorithm training and testing.
Figure 1: Word accuracy rates (OWER%) for algorithms across datasets.
Dataset
Metric
Deep Learning
CRF
Liang
Brandt Corstius*
CELEX
OWER%
0.446 (±0.038)
0.575 (±0.037)
1.460 (±0.073)
16.819 (±0.125)
OLER%
0.068 (±0.007)
0.087 (±0.005)
0.180 (±0.006)
3.143 (±0.014)
HER%
0.145 (±0.017)
0.178 (±0.016)
0.438 (±0.011)
6.818 (±0.023)
Precision
99.852 (±0.016)
99.807 (±0.011)
99.669 (±0.013)
93.350 (±0.037)
Recall
99.855 (±0.017)
99.822 (±0.016)
99.562 (±0.011)
93.182 (±0.023)
F1
99.853 (±0.015)
99.814 (±0.012)
99.616 (±0.012)
93.266 (±0.027)
Table 3: Accuracy means and standard deviations for metrics for algorithms tested across three datasets. Best performing algorithm for each metric is displayed in bold.
Figure 2: Architecture of model C, combining model A and B.
Figure 3: Word-level error rate after each epoch for the orthographic (blue), phonetic (red), and combined (green) models. The combined model, which integrates information from both the pretrained orthographic and phonetic models, shows a lower error rate on the validation set than the individual models.
Dictionary Words
Loanwords
Pseudowords
abonneren
mesjogge
frijk
vignet
internet
millen
keten
videoscherm
sjobel
ontleer
ciao
zwieten
dier
yakuza
wachteniek
bezorgd
telewerken
sprieuw
Table 4: Sample words from the three different datasets used.
Input
Model
Translation
Orthographic
Phonetic
Orthographic Only
Combined
suede
sy’w)d@
sue-de
su-e-de
suede
leptosoom
lEpto’som
lep-tos-oom
lep-to-soom
leptosome
aasgieren
’asxir@
aas-gi-e-ren
aas-gie-ren
vultures
hostesses
’hOst@s@s
hos-tes-ses
hos-tess-es
hostesses
care
’k)r
ca-re
care
care
Table 5: Comparison between output from the orthographic model (third column) and the output from the combined model (fourth column) that predicted correct syllable-boundaries through use of phonetic information (second column).
Syllable-level units offer compact and linguistically meaningful representations for spoken language modeling and unsupervised word discovery, but research on syllabification remains fragmented across disparate implementations, datasets, and evaluation protocols. We introduce findsylls, a modular, language-agnostic toolkit that unifies classical syllable detectors and end-to-end syllabifiers under a common interface for syllable segmentation, embedding extraction, and multi-granular evaluation. The toolkit implements and standardizes widely used methods (e.g., Sylber, VG-HuBERT) and allows their components to be recombined, enabling controlled comparisons of representations, algorithms, and token rates. We demonstrate findsylls on English and Spanish corpora and on new hand-annotated data from Kono, an underdocumented Central Mande language, illustrating how a single framework can support reproducible syllable-level experiments across both high-resource and under-resourced settings.
Héctor Javier Vázquez Martínez
Department of Linguistics, University of Pennsylvania, USA
The goal of circuit discovery is to decompose a monolithic neural network into tiny modular subnetworks which are interpretable in the sense that each does an identifiable subtask. We develop a practical method for composing arbitrarily many circuits into a single model. We propose a greedy, density-scaled composition algorithm that overcomes limitations of previous attempts to compose circuits by naive union, then generalize it to N circuits via CircuitTree, a new data structure built on a Θ(NlogN) tree-compose schedule. We compose circuits pruned from HuBERT for classifying individual phonemes into a syllable classifier, a capability none of the 39 phoneme circuits has individually. Chaining CircuitTrees into a hierarchy-of-circuits pipeline, we transcribe open syllables at 64.8% accuracy, within 1.2 points of a 66.0% ceiling that shares no circuitry at all, with 99.5% of circuit components shared.
German is a morphologically rich language whose syllable structure is exceptionally well-predicted by the Knuth--Liang hyphenation algorithm. We ask whether phonologically informed tokenization can serve as a competitive target for end-to-end speech recognition. We compare three tokenizer families on the Omnilingual ASR wav2vec 2.0 backbone fine-tuned with CTC: the pretrained multilingual character inventory, a data-driven Byte-Pair Encoding (BPE) over orthography, and phonologically informed units from Pyphen syllabification and grapheme-to-phoneme conversion. Across 40 fine-tunes, we evaluate on three German test sets spanning orthogonal shifts: in-domain read speech, dialectal spontaneous speech, and standard-German spontaneous speech. In-domain, all phonologically informed tokenizers match BPE and the multilingual character baseline on both WER and CER. Under domain shift the picture splits along vocabulary size rather than the linguistic axis of variation: at small vocabularies, syllable-aware tokenization improves on dialectal speech, where phonetic surface forms vary but syllable structure is preserved, and stays ahead on spontaneous speech, where new word-forms violate vocabulary closure. A phoneme-level confusion analysis further shows that all tokenizers commit the same canonical function-word errors, indicating that the acoustic encoder, not the tokenizer, dominates the error topology. Our findings suggest that tokenizer choice may depend on the vocabulary budget as much as on the distribution shift expected at deployment rather than reducing to a single universal optimum.
Christopher Witzl, Tobias Bocklet, Korbinian Riedhammer
Technische Hochschule Nürnberg Georg Simon Ohm, Nürnberg, Germany