11 April 2005 Class Notes on: Rice, K., & Avery, P. (1995). Variability in a deterministic model of language acquisition: A theory of segmental elaboration. In J. Archibald (Ed.), Phonological acquisition and phonological theory (pp. 23-42). Hillsdale, NJ: Lawrence Erlbaum. Kirk Baker p.27 I have a vague idea of what functional load is, just wondered if the concept is quantifiable or if it's meant as a not precisely defined but intuitively appealing explanation in general. ***Functional load as a measure of the amount of work a contrast does in a language -- could be quantified by counting e.g. the number of minimal pairs distinguished by a given contrast. Different from phonotactic probabilities in that functional load requires comparisons of multiple values? (Is this right?) For example, a language whose lexicon consisted of [ka] [ke] [ki] [ko] [ku] has a high frequency of [k], which might influence acquisition of [k], but [k] has a low functional load since no contrasts are made with it. But a language with [ka] [ke] [ki] [ko] [ku] [pa] [pe] [pi] [po] [pu] has a higher functional load on the place of articulation value for consonants. p.27 Why do they say a probabilistic model "predicts substantial individual differences". Why do they say a prob model "makes no empirical predictions"? ***Differences come from child's selection of elements from the environment? But how much variation does there have to be to have "substantial" variation? How do we know if there is "enough" variation to count as substantial and hence uphold the theory's predictions? Are they being hypocrtical to say that the cognitive model is the wrong way to go because of the variability it predicts, only to use the child's selection of a particular path along the pre-determined phonological structure to explain variability? p.31 Not sure what they mean by dependency. Seems like SV would depend on Airflow at least articulatorily. p.34 Discussion of [p,t,k] as phonetic realization of same phoneme...first distinction should be /p/ vs. /t/..."in general borne out" based on F&F 1975's study of _3_ children?! ***Yes, we all agreed that more data needs to be brought to bear on their claims! p.37 Table 2.1 C-coda is most variable in child's production. This position also shows the most noncontrastive variability in adult speech. p.40-41 Would be nice to see actual data illustrating their theory. p. 42 conclusion. "small number of different structural paths made available to the child". What determines the path any child takes? By any chance probabilistic? Didn't really buy their arguments against a probabilistic model, not sure their model is completely deterministic, not sure I "wonder about the validity of adopting a probabilistic cognitive approach" Junko Davis In 4.3. (p.41) they discuss individual variability. I do not quite understand their account of the individual variability related to favorite sounds. Why is it not phonologically important? ***Our guess is that Avery and Rice are using a very traditional view of "phonology" to mean mental representations of sound structure, isolated from other environmental factors. So while practicing a favourite sound may increase a child's ability to produce that sound accurately and influence acquisition, it does not affect the phonological representation of the sound or its relation to other sounds in the system. "Although frequency of a sound in an ambient language may well have an effect, it does not appear to override the general constrains placed by our model." (p.41) -- How do they account for the reported systematic differences? ***They don't. I could have been more convinced if the arguments were accompanied by more actual data. Kathleen Currie Hall Summary: In this paper, Rice and Avery present a theory of phonological acquisition that relies on unary, privative features and markedness. They claim that phonological acquisition will proceed along certain pre-determined paths, defined by phonological features that define sounds in terms of contrast to other sounds. Along any given path, there is only one route (which they say explains "universals" in acquisition, but a child may follow different paths (hence variability in acquisition). Details: 1. History: competition between models of acquisition . . . a. Full specification of features universal; as learn language, lose specifications of redundant info. b. Minimal specification of features; as learn language, add specifications as needed. R&A: minimal spec. c. Universal, deterministic approach (Jakobson) d. Cognitive, variable appraoch (Ferguson) R&A: deterministic with variation 2. They claim that there are two types of sounds (basic/repertoire sounds, and complex/non-repertoire sounds). Children start with basic sounds and then add on complex ones -- this is one of the universal patterns to be accounted for. 3. Their proposal is that there is a universal phonologcial representation for sounds, resembling a feature geometric model. The child starts out with an impoverished version of this, and as they learn new phonological contrasts, structure is built up. Because the structure has root nodes and expanded daughters, they can claim that addition of structural information along a given path is universal and deterministic; however, choices between nodes or paths to follow can introduce variability into the system. Questions / Comments: 1. p. 27 -- they dismiss functional load out of hand, saying that "the child would appear to require some foreknowledge of the total inventory of the language in order to determine the functional load of individual segments" -- but this isn't true, is it? I would have thought that it isn't that the child thinks "oh, that has a high functional load, I'll learn it first" but rather that the child will naturally be drawn to learning contrasts first that do a lot of work in the language (because they are frequent and useful, etc.). ***Yes, we think that it will be the role that the sounds with high functional load play in the language that will have an effect, regardless of the child's knowledge of the adult language system as a whole. Also bear in mind that what is probably relevant is the functional load of contrasts in the language that the child _has_ heard up to whatever point is relevant for making phonological decisions, which the child certainly does have access to! 2. Note that they assume a universal markedness theory -- cf. e.g. Hume and Tserdanelis (2002) about labial unmarkedness in Sri Lankan Portuguese Creole and other evidence of language-specific markedness. ***Note also guidelines for "typical development" -- by age 3, should have [p, b, m, n, w, h]; not until age 5 is [t] necessarily acquired. Does this say anything about markedness? 3. Is citing 3 kids here and there, presumably from the same language within each study, really convincing evidence for their theory?! 4. I feel as though they're saying that the structure is universal, but especially in the beginning, there is no way to see this because any phonetic realization counts as a manifestation of the universal structure -- so this is an unfalsifiable claim. 5. How do you tell the difference between a child who produces his single "placeless" consonant in mulitple places (p,t,k) and one who actually has a place contrast? Is it only evident once the child clearly has meaningful words? Is this necessarily when "phonology" starts? (see also p. 37 in the table -- how do you know there's a place distinction at the beginning of the word and not at the end? what about variable input and context effects? especially bad example since the final C is a coronal and we know it undergoes assimilation . . . 6. p. 35, 39 -- why do they say that the first contrast is between coronal and non-coronal, if they mean coronal and labial? or do they predict that the first (place) contrast will be between t/d/n on the one hand and a sort of super-consonant p/b/m/k/g/N on the other? 7. So how on earth do they explain any patterns that aren't deterministic (and surely there are some)? 8. And how do they account for universal patterns even within their deterministic models -- e.g. that kids pick the "place" path or the "manner" path earlier than the "SV" path -- e.g. p. 41 "this seems to be a less common elaboration" -- can they explain these kinds of effects? 9. p. 41-42 "Although frequency of a sound in an ambient language may well have an effect, it does not appear to override the general constraints placed on our model" -- what do they base this claim on? are there no cases in which children get a contrast between p/k before p/t or even t/k? Besides, even if it is not incompatible with their model, their model has no way of accounting for or representing such patterns. Why not? Do they not consider this part of phonology proper? Eunjong Kong Minimal specification approach to the acqusition of segmental representation claims that segments in the child's inventory are stored in a way of being constrative to each other. The increase of contrast adds more complex structured segments into the inventory. - In this theory, unmarked segments are represented with 4 major nodes of internal organization which are underspecified. Additional specifications under each node represent the contrast among segments added into the inventory. The system develops by way of increasing these segmental contrasts (elaboration). In acquistion, the theory supposes the universal path by which the contrasts is added into the system, which determines which node must be specified earlier than others.(coronal-> periperal-> labial or dorsal...) - The lack of contrast(not acquired yet) produces variable realizations of the same segment(non-contrastive variability). - The order of specification introduced under each node results in many different possible path that the child can take (structural variability). - Individual variability is attributed to the difference of child's favorite sounds and frequency effect of the ambient language. Questions: - This theory seems to assume that infants'speech production directly reflects their phonological representation of segments(contrast structure). If so, would it be a reasonable assumption? ***What is even meant by "phonological representation"? (This came up even more in the discussion of Dinnsen and Chin.) What about articulatory ease? Also visual feedback -- big cue for labials? Random choice? Selectional attention of the child? Fangfang Li This paper tries to account for both universality and variability in language acquisition using a deterministic model and a minimalist approach. Basically, it claims that universality can be found by a relatively fixed path specified in a model of markedness and underspecification. Also since there is no constraint on the four parallel organizing nodes (Laryngeal, Air flow, SV and places), a certain degree of variability is allowed in the overall picture. It supports Jocabson's view that children acquire contrasts instead of individual segments. Questions: 1. (pp. 29) It mentioned Locke (1983)'s distinction of "repertoire sounds and nonrepertoire sounds", with repertoire sounds as the basic sounds found in children's early sounds as well as the most frequent sounds across languages in the world. What sounds are included here? Consonant or vowels or both? If they talk about sounds as phonemes, then different language would have different phonemes with different analysis methods. How can they determine the common sounds shared by most of the languages then? 2. (pp 31) "The unmarked features are default features added for the purpose of phonetic implementation and, in general, do not play a role in the phonology". What do they mean by "phonetic implementation here?" Will this phonetic implementation varies from language to language? The thing I have in mind is that English, Japanese, and Mandarin Chinese all have the sound notated as sh, but they are phonetically very different from one and another. Is this something that would fall into "phonetic implemenation" that Rice and Avery talked about here? ***Yes, these types of differences would probably go into the "phonetic" variability component of the model? The main point is that there is a single internal mental representation of the contrasts between these sounds, which all children will eventually acquire. 3. (pp. 33) How native English speakers would pronounce "bike belt"? Could [k] in bike be assimilated to the followed labial sound? If yes, it will be contrary to their hypothesis that coronal is unmarked. ***No, the English speakers all agree that *[baibbElt] is a bad pronunciation of this. So this wouldn't be counter-evidence, though we feel sure that there is such evidence in general. 4. (pp. 37, Table 2.1) This Tessa's case does not show their point in my eyes, if I did not interpret it wrong. When it talks about noncontrastive variability, I understand it as noncontrastive in the same position over all the words that the child has. But in Tessa's case, it only talks about the word "boat", and did not mention whether Tessa has other words that ends with aother consonant or not. According to their theory, unless "boat" is the only word with consonant final that Tessa knows, but it seems unlikely to be the case if by age 1:9:2, Tessa did not acquire other similar words. ***Point very well taken. It is impossible to determine anything conclusively about Tessa's representations when we don't have any other words! How do we tell the difference between her acquisition of a contrast and the presence or absence of more or less variability? What about variability in the adult input? Ila Nagar 1. The main point of the article is to advocate a segmental approach to language acquisition. The authors focus on the way in which children acquire individual sounds and how they distinguish one sound from another. They follow the minimal specification approach which they explain in a very convincing manner. 2. I found the article confusing when the authors started explaining the segment structure and then tried t put the children learn within that structure. Their arguments were convincing but their methodology was not. In section 4.1, they make a point which in some way tied their earlier discussion with their main point. 3. Vihman and Rice and Avery take very different approaches to child language acquisition. While Vihman focuses on individual variation, Avery and Rice said their proposal could explain it. I would point that as a short coming in Rice and Avery?s work. They also do not rely as much on data as they should for this kind of study. Helen Riha p. 24 “The child is guided in the construction of linguistic representations by invariant principles of UG, as well as by a set of tightly constrained open parameters that are fixed by experience.” As we have seen from reading other work on language acquisition, the notion of “experience” needs to be defined. We have seen that there are specific ways in which children obtain language experience, especially via interaction with their mothers and also by imitation. Rice and Avery do not address the issue of experience, but an awareness of types of experience could affect their theory of phonological development. 
***Yes, the roles of environment and of motivation seems to be vastly underused when it comes to formulating linguistic theories. Children aren't in a vacuum; they pick things up very quickly, especially if they are motivated; how do these shape phonological acquisition? 
p. 24 “We claim that one expects to find a good deal of variation in the realization of sounds when the segment structure is relatively unelaborated.” Isn’t some of this variation also due to physical constraints on motor movements in articulation and in children’s ability to imitate what they hear? 
***Yes, this is only briefly mentioned in the article. 
p. 30 “We assume … that segments have internal structure and that features are grouped together under higher level organizing nodes.” It seems that in acquiring phonology, children build their phonological inventory based on the sounds and larger units they know and those they recognize in the speech of others -- their data is limited. Wouldn’t this affect the accuracy with which they make assessments about segment structure? Would children necessarily realize that segments have the internal structure and feature organization shown in (1)? 
***Good question. Is there any reason that all children actually _do_ have to come up with the same final internal mental representations? Could there be multiple ways of representing the same structures? How would we know? See also Dresher's Successive Division Algorithm in which the same phonological structure could be arrived at differently.
 p. 30 Rice and Avery argue that children move from impoverished phonological representations to fully elaborated representations. They do not account for the fact that children also make erroneous judgements about structure during acquisition (e.g., creating idiomorphs like yesbody and hi-ed daddy) and subsequently revise their understanding of structure. How would Rice and Avery account for these erroneous judgements and their later revision? How do wrong judgements about structure fit into their model? 
p. 34 “When acquiring a sound system, the child begins from an impoverished structure with a wide range of possible realizations for any particular segment.” Could it be the case instead that children acquire larger units like words or syllables with an impoverished structure? In other words, taking Jakobson’s example of [p] and [a] from very early language acquisition when children have a distinction only between consonants and vowels (p. 34), could it be the case that they have an impoverished structure of [pa] as a unit rather than [p] and [a] as separate units? At the heart of this issue is the question of which units are really relevant to the child during early language acquisition. Are individual segments meaningful in themselves, or are they only meaningful when embedded in larger units?
 ***Think also about adult language: studies like Goldinger&Azuma show that adults can selectively pay attention to particular units. Why not children? Is it also based on personal bias? How do such choices shape acquisition? Are segments even really relevant to children? At what point? What about evidence that words are acquired first? Vs. systematic differences from adult outputs that seem to pinpoint segments? (see also Dinnsen and Chin discussion) Laura Slocum Brief Summary: Rice & Avery explore theories of child language and phonological acquisition, using some of Jakobson's (1941/1968) observations as the starting point for discussion. These authors serve to defend several of Jakobson's claims, including determinism and universality, by attempting to sort out what they term the "variability problem." The basic premis is that children initially produce basic contrasts between or among a select set of sounds, which are maximally distinctive. Therefore, differences within and across languages tend to be less when inventories are small. As children's productions become constrained by the phonological rules common to their ambient language: 1) individual inventories increase in size, and 2) variation within and across languages increases. Among the factors which are thought to impact individual variability in acquisition are sound preference and frequency effect. Comments and Questions: At a recent Speech-Language-Hearing convention, the idea of using "maximally contrastive" phonological therapy was re-visited in order to promote more global change in the child's phonological system. I wonder if it is necessarily the type of therapy and perhaps not just the intensity of therapy which effects change in (disordered) children's speech and language skills. What is the proposed mechanism for change in a "maximally-contrastive" approach? ***Hmm, we didn't cover this question directly, though related questions came up in the Dinnsen and Chin discussion. But more generally, how does theory like that proposed by Rice and Avery affect child language training / speech pathology? How do we know if the proposed model is correct? What do we do to reconcile it with what we consider "typical development" stages? How do we go about making sure the child is making the "right" assumptions about phonological structure if these really are what is required for correct acquisition? Asimina Syrika Comments This article looks at the acquisition of inventories and examines the consequences of segmental structure and underspecification. 2 approaches to the acquisition of segmental representations: -Full specification approach: Segments are initially acquired as units. The child begins with a universal representation of a sound that is fully specified and then as the child's inventory of sounds becomes more elaborate these initial fully specified representations may be modified. This happens through the deletion of redundant features as the contrasts are acquired. -Minimal specification approach: At any time the child has only enough specification to keep a given segment distinct from the other contrastive segments in the inventory. Thus, if the child consistently used a sound like /p/, we would know little about the child's representation of the sound without knowing what other segments were in the inventory. Rice and Avery take the minimal specification approach and view the acquisition of representations as the elaboration of segment structure (through addition of contrasts) along a predetermined pathway. They claim that a lot of variation in the realisation of sounds can be found in the early stages of language acquisition when the segment structure is relatively unelaborated, but as the segment structure becomes more complex (through the acquisition of more contrasts) the phonetic realisation of sounds becomes less variable. More complex segments are also acquired as contrasts are added to the child's phonological system. Two theories relevant to Rice and Avery's model of acquisition: -Segment structure (segments have internal structure, features are grouped together under higher level organising nodes, dependency relationships exist between nodes). -Underspecification (unmarked material, redundant feature values and predictable material are underspecified). Two central hypotheses to their theory: a. Minimality: Initially the child has impoverished (minimal) structure. b. Monotonicity: Inventories are built up in a monotonic fashion (starting point is set at completely unmarked structure). Importance of elaboration The acquisition of new contrasts adds complexity to segment structure. Accounting for variability in this model of elaboration: -Noncontrastive variability (variability from absence of structure) -Structural variability (Variability depending on the order in which the different nodes are elaborated) -Individual variability Questions 1) What is the theory of underspecification and how does it relate to language acquisition? ***The basic idea is that the child starts with no phonological structure and then builds up contrasts as they become necessary for representing the phonological contrasts in the language. Rice and Avery assume that all children will build the same structures and hence they make particular claims about the order of acquisition and the implications of having particular contrasts on the structure of the language as a whole. 2) Rice and Avery mention two central hypotheses to their deterministic model of language acquisition: minimality and monotonicity. I was wondering how universal monotonicity is. How do they account for cases of exception to this 'only pathway' to the addition of structure in the child's phonology? How do they account for cross-linguistic differences in the order of acquisition and mastery of individual phonemes? ***Again, they don't seem to have any way of doing this. 3) Rice and Avery base their model on many theoretical assumptions. In the absence of empirical data, how do you know what is the underlying representation of a child? What constitutes an 'impoverished underlying representation'? Does it only have to do with lack of systematicity in the phonetic realisation of sounds? ***This is very unclear in their model. See also discussion from Dinnsen and Chin. 4) I don't think they really consider variability in their model. They don't mention input effects, either, but they rather see 'no phonological relevance to the choice of sounds' (p.41) by the child. They say that the child has 'a small number of different structural paths' (p.42) available. What determines the kind of path, though? Is it completely random or does it have to do with biological constraints, input or frequency effects?