Liquid Deletion in French Child-Directed Speech

Sharon Peperkamp, Monica Hegde, Maria Julia CarbajalInterspeech 2019

TL;DR

This paper investigates "liquid deletion" (e.g., saying "tab" instead of "table") in French child-directed speech. They found this phonological rule is surprisingly common and that instances of words with and without the deletion tend to cluster together, suggesting a potential learning mechanism for young children.

What Problem Does This Paper Address?

Young children learning a language must not only learn to produce words correctly but also understand variations in pronunciation. This includes understanding phonological rules, which are systematic changes to sounds in speech. One such rule is "liquid deletion," where sounds like 'l' or 'r' are dropped in certain contexts, particularly at the end of words. While children seem to develop the ability to "undo" these rules early on for comprehension, how they learn them is still a mystery. This paper tackles this by examining the input children receive: specifically, how often liquid deletion occurs in French child-directed speech and whether the way these variations appear might help children learn the rule.

Methods

The researchers analyzed a corpus of spontaneous French interactions between parents and their young children, recorded at home. They specifically focused on utterances spoken by adults. Using video recordings and audio, they identified words ending in obstruent-liquid clusters and coded whether the liquid sound was deleted or present. They also noted the phonological context (before a consonant, vowel, or pause) and the speaker. To investigate potential learning cues, they analyzed the temporal distance between occurrences of the same word with and without liquid deletion. They also compared this clustering to "true phonological neighbors" – words that differ by a rule but are not necessarily the same lemma.

Datasets

The study utilized the Yamaguchi corpus from the Childes database, which contains video recordings of spontaneous interactions between a single child and their parents, as well as other adults. Twenty-six recording sessions were selected, covering children aged between approximately 1 and 4 years.

Key Insights and Ideas

  • Liquid deletion is common in child-directed speech: Contrary to a potential assumption that parents might simplify speech less for children, liquid deletion is a frequent occurrence.
  • Clustering of rule-varied words as a learning cue: The finding that words with and without liquid deletion often appear close together in speech is a significant insight. This temporal proximity could act as a "signpost" for children, drawing their attention to the phonological variation and aiding in rule acquisition.
  • Distinction from true phonological neighbors: The clustering of rule-varied tokens of the same word is more pronounced than the clustering of different words that are phonologically similar but not related by the same rule. This suggests the clustering is specifically tied to the phonological alternation itself.
  • Potential for learning without meaning: The authors propose that this clustering mechanism might allow infants to acquire the rule of liquid deletion before they even know the meaning of the words involved, offering a novel perspective on early phonological learning.

Key Results

  • Liquid deletion occurred in over half (58.0%) of the analyzed word tokens.
  • Liquid deletion was more frequent in the mother's speech (65.7%) compared to the father's (53.4%) and the researcher's (55.3%).
  • Deletion was significantly more common before consonants (79.1%) than before vowels (29.9%) or pauses (19.8%).
  • Instances of the same word with and without liquid deletion occurred with a median distance of 49 seconds (or 12 utterances).
  • Pairs of true phonological neighbors (different words differing by a rule) had a significantly larger median distance of 39 utterances, indicating less tight clustering.

Limitations and Future Work

  • Limited direct comparison to adult-directed speech: The corpus primarily contained child-directed speech, making a direct comparison to adult-directed speech difficult. While the authors ran analyses excluding adult-directed tokens and found similar results, a more robust comparison would require dedicated adult-directed data from the same speakers.
  • Speaker variation: While differences were noted between speakers (mother vs. father vs. researcher), the authors suggest more data is needed to definitively conclude on gender or other speaker-based variations in liquid deletion.
  • Future research directions: The authors propose testing whether infants are indeed sensitive to this clustering phenomenon and at what age they acquire liquid deletion and begin to compensate for it during sentence processing. Further research could also explore how this mechanism might apply to other phonological rules.

Summary updated

BibTeX
@inproceedings{peperkamp2019liquid,
  author = {Sharon Peperkamp and Monica Hegde and Maria Julia Carbajal},
  title = {{Liquid Deletion in French Child-Directed Speech}},
  booktitle = {Interspeech},
  year = {2019},
}

Loading related content...

Figure