Periodic Reporting for period 1 - SoundAct (The actuation of sound change)
Reporting period: 2023-02-01 to 2025-07-31
Historical sound change has its origins in the different ways in which sounds are transmitted between a speaker and hearer in everyday conversation. But under what conditions can such variation lead to sound change? This question is part of the actuation of sound change, recognised as one of the greatest challenges in linguistics, and which is about the conditions under which sound change takes place, and why languages can follow such different paths of sound change. The actuation puzzle remains unsolved principally because the beginning of sound change is so gradual that it is undetectable even with modern instrumentation. The project remedies this deficiency by recasting the elusive actuation puzzle through analyses of two dialects, A and B of the same language, such that sound change is more advanced in B than in A. To achieve generalisation, the dialect pairs are chosen from Bantu, Indo-European, and Japanese languages that differ markedly in their sound patterns and sociocultural background. The degree of A→B mapping is determined by the application of speech processing techniques combined with agent-based modelling in which real speakers from the dialects are represented by computational agents that communicate with each other. The combination of these analyses of real and simulated speech data are used to model the relationship between speech communication and sound change as a complex system in which speech sounds that are copied from each other by individuals in everyday conversation sporadically give rise to a shift in the pronunciation pattern of a dialect that affects the entire community.
The primary objective is to explain why sound change should happen under one set of circumstances but not another i.e. to account for the set of factors that separates language stability from language change. The associated secondary objectives are (1) to understand how very slight and often random variation in the speech signals transmitted between speakers and hearers can cause a grammatical, categorical sound change in the sound pattern of the dialect (2) to determine whether such a grammatical change is precipitated by certain types of social network structures that bind individuals in a community and (3) to make use of the cognitively inspired, agent-based, computational, model of speech to test what types of mapping between words, their pronunciation patterns, and the physical speech signals are most likely to lead to sound change.
The primary objective is to explain why sound change should happen under one set of circumstances but not another i.e. to account for the set of factors that separates language stability from language change. The associated secondary objectives are (1) to understand how very slight and often random variation in the speech signals transmitted between speakers and hearers can cause a grammatical, categorical sound change in the sound pattern of the dialect (2) to determine whether such a grammatical change is precipitated by certain types of social network structures that bind individuals in a community and (3) to make use of the cognitively inspired, agent-based, computational, model of speech to test what types of mapping between words, their pronunciation patterns, and the physical speech signals are most likely to lead to sound change.
We completed field trips and recorded speech data from three of the four languages targeted for analysis. These were from speakers of (a) three dialects of Italo-Romance in Southern Italy (b) three dialects from the Meru region of Kenya and (c)f the Kagoshima region of Southern Japan and of the standard Tokyo dialect. These recorded speech data were semi-automatically annotated and organised into speech databases using existing tools at the host institute for the purposes of the subsequent empirical analysis. An additional speech database was compiled of existing speech physiology data of the movement of the vocal organs obtained from real-time magnetic resonance imaging of two English dialects.
Software modifications and testing were made to the agent-based computational model that has been developed at the host institute over a number of years in order to adapt it to processing differences in word morphology (e.g. the difference between singular and plural or gender) that form a central part of the sound changes under investigation.
The main achievements are in (i-iv) below. These are with respect to the secondary objectives (1) categorization and (3) the architecture of the cognitively inspired computational model of sound change. The analysis of (2) social network connections is targeted for the latter part of the project after data from a larger number of participants has been collected and processed.
(i) The project has been able to verify from analyses of real speech recordings from English, Italo-Romance, and Meru that there are dialects that are positioned at more conservative (A) and innovative (B) stages of sound change.
(ii) The analysis of the data from the physiological, MRI study of Southern Standard British (A) and American (B) English has shown how variation in the soft-palate and tongue tip is at the origin of the development of sound change by which vowels are nasalized and final nasal consonants are lost (e.g. French 'main', /mɛ̃/ from Latin 'manus' (hand).
(iii) The acoustic data from Italo-Romance has been used to explain the origin of sound changes by which morphological information is transferred from a suffix vowel (e.g. Standard Italian 'mese', 'mesi') to the stem with loss of the suffix (/mes, mis/, 'month', 'months' in the most innovative dialect).
(iv) The agent-based computational model of sound change has been adapted to be able to model morphologically-based sound change of the kind in (3) above.
Software modifications and testing were made to the agent-based computational model that has been developed at the host institute over a number of years in order to adapt it to processing differences in word morphology (e.g. the difference between singular and plural or gender) that form a central part of the sound changes under investigation.
The main achievements are in (i-iv) below. These are with respect to the secondary objectives (1) categorization and (3) the architecture of the cognitively inspired computational model of sound change. The analysis of (2) social network connections is targeted for the latter part of the project after data from a larger number of participants has been collected and processed.
(i) The project has been able to verify from analyses of real speech recordings from English, Italo-Romance, and Meru that there are dialects that are positioned at more conservative (A) and innovative (B) stages of sound change.
(ii) The analysis of the data from the physiological, MRI study of Southern Standard British (A) and American (B) English has shown how variation in the soft-palate and tongue tip is at the origin of the development of sound change by which vowels are nasalized and final nasal consonants are lost (e.g. French 'main', /mɛ̃/ from Latin 'manus' (hand).
(iii) The acoustic data from Italo-Romance has been used to explain the origin of sound changes by which morphological information is transferred from a suffix vowel (e.g. Standard Italian 'mese', 'mesi') to the stem with loss of the suffix (/mes, mis/, 'month', 'months' in the most innovative dialect).
(iv) The agent-based computational model of sound change has been adapted to be able to model morphologically-based sound change of the kind in (3) above.
The potential impacts are in at least:
1. Complex systems, by using an agent-based model to show how change to the grammar i.e. pattern of speech sounds that characterises a community is affected by interlocutors' production, perception, and memorization of speech signals.
2. Dialectology, through the first ever acoustic analyses of dialects from Italo-Romance and Meru (Kenya) as well as the first large-scale physiological comparison of nasalization in two English dialects for understanding sound change.
3. Historical linguistics, through acoustic and physiological analyses that shed light on sound changes including vowel nasalization and vowel metaphony/umlaut found in many unrelated languages.
4. Human models of speech processing, because the analyses contribute to explaining how consonants and vowels that form the basis of meaning distinctions between words are related to the production and perception of speech signals in time.
The further success of the project depends on at least the following factors:
1. Finding pairs of dialects of the same language that differ from each other in such a way that they provide useful information for understanding how phonetic variation and sound change are connected.
2. Being able to record, process, and analyse speech from such dialects rapidly and for a much larger number of participants than hitherto analysed. This is essential for being able to address successfully secondary objective 2 concerned with the influence of social ties and network structure on sound change.
3. Establishing which combinations of parameter settings of the agent-based computational model of sound change are able to map between dialects that are less and more advanced in sound change.
1. Complex systems, by using an agent-based model to show how change to the grammar i.e. pattern of speech sounds that characterises a community is affected by interlocutors' production, perception, and memorization of speech signals.
2. Dialectology, through the first ever acoustic analyses of dialects from Italo-Romance and Meru (Kenya) as well as the first large-scale physiological comparison of nasalization in two English dialects for understanding sound change.
3. Historical linguistics, through acoustic and physiological analyses that shed light on sound changes including vowel nasalization and vowel metaphony/umlaut found in many unrelated languages.
4. Human models of speech processing, because the analyses contribute to explaining how consonants and vowels that form the basis of meaning distinctions between words are related to the production and perception of speech signals in time.
The further success of the project depends on at least the following factors:
1. Finding pairs of dialects of the same language that differ from each other in such a way that they provide useful information for understanding how phonetic variation and sound change are connected.
2. Being able to record, process, and analyse speech from such dialects rapidly and for a much larger number of participants than hitherto analysed. This is essential for being able to address successfully secondary objective 2 concerned with the influence of social ties and network structure on sound change.
3. Establishing which combinations of parameter settings of the agent-based computational model of sound change are able to map between dialects that are less and more advanced in sound change.