Segmental and Prosodic Evidence for Property-by-Property Transfer in L3 English in Northern Africa
Marit Westergaard
Round 1
Reviewer 1 Report
Review of “Segmental and Prosodic Evidence for Property-by-Property Transfer in L3 English in Northern Africa” for Languages, December 2021
Intro
This is a well written article on a central topic in L3 acquisition, providing a theoretically very interesting analysis of data from understudied languages. The empirical data come from previous studies of L1 Arabic–L2 French–L3 English. The paper discusses the notoriously difficult concept of structural similarity between languages and introduces a very promising notion of I-proximity. The article is of high quality and clearly suitable for the special issue.
Major comments
Data
The analysis in the paper is not based on a recent study carried out by the author; instead, the data analyzed are taken from old studies published in Benrabeh (1991), which focuses on the pronunciation of vowels and consonants, and Ghazali (1973) and Ghazali & Bouchhioua (2003), focusing on word and sentence stress in L3 English. This should be mentioned already in the abstract and the introductory section of the paper (as it is now, it comes as a bit of surprise in the Results section on pp. 4-5). The background information about the studies is extremely sparse; we are only told that the participants are (high school) students who have acquired French for 8 and English for 3 years. However, to the extent it is possible, all relevant background information about the studies and the participants should be provided (number of participants, age, methodology, size of corpus, how are the data recorded and transcribed, etc.) (there are two paragraphs of background information in the Results section on p. 6 – this should be moved to an earlier section).
Another important issue for current L3 studies is to provide information on the participants’ proficiency in the previously acquired languages, and minimally, the properties under investigation should be tested in both languages in order to make sure that they have been acquired properly in the L2 (and that they have not attrited in the L1). For the current studies, the most interesting information of this kind would be to know to what extent the participants’ L2 French is influenced by their L1 Arabic. In case the relevant French properties (consonants or word stress) have not been target-consistently acquired (but has many features of Arabic), then what may seem like property-by-property transfer from both languages may actually be wholesale transfer from a French L2 containing L1 Arabic features (which would be predicted by the TPM). Thus, to the extent that there is any information available on the L2 French of these participants, this should be provided in the paper. And if this information is not available, the author should at least mention and discuss this.
The LPM
On p. 8, the author writes the following: “Westergaard’s (2021) LPM draws on the parsing models proposed in previous works such as Fodor (1998a, 1998b) and Lightfoot (2020).” However, the LPM was developed in 2014 (and e.g. presented at BU that year; see Mykhaylyk et al. 2015), so could not have been inspired by Lightfoot (2020). The LPM is an extension of the Micro-cue Model (MCM, see e.g. Westergaard 2008, 2009a,b, 2013, 2014), to multilingual situations (see e.g. Westergaard 2019, section 2). The MCM argues for learning by parsing and incremental development in small steps, based on input/intake, a reduced UG (without parameters) and 3rd factors (e.g. economy). Thus, the MCM is inspired by Fodor’s work as well as Lightfoot’s traditional work on cue-based acquisition and change (Lightfoot 1999, 2006). The model in Lightfoot (2020) is very similar to the MCM but seems to have been developed independently.
The Tolerance Principle
The author devotes a short section to the Tolerance Principle (TP), arguing that it provides the mechanics of comparison between different selections for parsing (the L1 or the L2). However, it is unclear how the TP could be used for this. The TP algorithm explains how learners formulate rules (vs. having to learn in an item-based fashion), and the number of exceptions that can be tolerated in this process. Deciding to use a parse from one language over another does not seem to be the same process as learning a rule; nor does it seem to be the case that leaving a number of the target forms unaccounted for would be considered exceptions to a rule. Furthermore, this would require the learner to somehow know in advance exactly how many segments the L3 has (and that have not been learned yet), cf. also the next section.
Furthermore, even if it were the case that the TP could be applied here, there would still be cases where it would not be able to decide between parses: Say the nature of the L3 system is such that parsing it with the L1 would leave 4 segments unaccounted for (and the TP would tolerate 5), while parsing it with the L2 would leave out 2 segments. In both cases, the TP would formulate a “rule”, but the learner would still have to make a choice between the two parses.
How big is a property?
The author shows (rather convincingly, in my view) how parsing selection can be affected by considering the inventory of segments in the three languages, rather than comparing segment by segment. However, learners presumably do not have a clear overview of segments in the L3 at an early stage and do not know exactly how many segments would be ambiguously represented if they choose to parse the L3 with the L1 vs. the L2. Thus, there is presumably a way to explain this by assuming that learners start out with a smaller inventory (say the most frequent or least complex or most similar segments); cf. the TP which argues that it is easier to formulate rules based on small inventories/lexicons.
There is a brief discussion on p. 14 whether property-by-property transfer could apply to smaller elements than vowels or consonants, e.g. obstruents or sonorants. I think this is a very important discussion and I suggest that it be elaborated somewhat here. This might also be related to the MCM, which argues that learners are sensitive to very fine/small distinctions in language acquisition, and the LPM, which argues that this sensitivity is also present in the process of crosslinguistic influence.
Full Transfer in L2A
The author seems to accept Full Transfer in L2A, but property-by-property transfer in L3A (according to e.g. item 3 in the conclusion). However, it is unclear whether full transfer in L2A means wholesale transfer, as it has been pointed out by e.g. Stringer (2021) that there has not been general agreement about this in the field of L2A. The field of L3 studies has brought this issue back to life, and it seems pertinent, given the strong position taken by the author for L3A, that the author clarify to readers exactly what is meant by full transfer in L2A, especially since it is also claimed that “there is unity if L2A and L3A”.
Minor issues
The paper is well written, but there are a few typos here and there, e.g. the following:
langauge -> language (pp. 1, 2, 10)
tri-lingual -> trilingual (p. 2)
properteis -> properties (p. 5)
envrionment -> environment (p. 7)
falifiability -> falsifiability (p. 8)
features structure -> feature structure (p. 12)
Note, that … -> Note that … (p. 13)
Princple -> Principle (p. 14)
On p. 3, there is a reference to Westergaard et al. (2019), which is not in the bibliography. – Perhaps 2017 is meant?
Some awkward formulations, e.g. on p. 5: A syllable with both a long vowel and which has a coda consonant is referred to as super-heavy.
The paragraph starting at the bottom of p. 6 (“The same results …) seems to be misplaced – as it is not well integrated with the previous paragraph. It is also much too short and should be elaborated.
The examples on p. 7 “the speaker asked for questions” vs. “she asked four questions” should presumably have identical subjects?
There seem to be some commas missing in the list provided in the Conclusions. Also, at the very beginning of the conclusion, “such an approach” has unclear reference.
References
Lightfoot, David. 1999. The development of language: Acquisition, change and evolution. Malden, MA and Oxford: Blackwell.
Lightfoot, David. 2006. How new languages emerge. Cambridge University Press.
Mykhaylyk, Roksolana, Natalia Mitrofanova, Yulia Rodina & Marit Westergaard. 2015. The Linguistic Proximity Model: The Case of Verb-Second Revisited. In Elisabeth Grillo & Kyle Jepsen (eds.), Proceedings of the 39th Boston University Conference on Language Development 2, 337-349. Somerville, MA: Cascadilla Press.
Stringer, David. 2021. Commitment in L3 relationships: Sacred vows of polyamory? Commentary to the Westergaard keynote. Second Language Research.
Westergaard, Marit. 2013. The Acquisition of Linguistic Variation: Parameters vs. Micro-cues. In Terje Lohndal (ed.), In Search of Universal Grammar: From Old Norse to Zoque, 275-298. Amsterdam: John Benjamins.
Westergaard, Marit. 2014. Linguistic variation and micro-cues in first language acquisition. Linguistic Variation, 14.1, 26-45 (post-print version).
Westergaard, Marit. 2009a. The Acquisition of Word Order: Micro-cues, Information Structure and Economy. Amsterdam: John Benjamins.
Westergaard, Marit. 2009b. Usage-based vs. Rule-based Learning: The Acquisition of Word Order in Wh-Questions in English and Norwegian. Journal of Child Language 36.5, 1023-1051. DOI: 10.1017/S035000909009349. Post-print version here.
Westergaard, Marit. 2008. Acquisition and Change: On the Robustness of the Triggering Experience for Word Order Cues. Lingua, 118.12, 1841-1863.
December 30, 2021
Reviewer[Name omitted]
Comments for author File: 
Comments.pdf
Author Response
Please see attached.
Author Response File: Author Response.pdf
Reviewer 2 Report
Summary:
This paper argues that L3 acquisition is influenced by both the L1 and L2 grammars on a property-by-property basis, depending on the similarity between the L3 and L1/L2 with respect to specific structures. This is in contrast to proposals, such as the Typological Primacy Model (Rothman et al. 2019), which propose that all aspects of either the L1 or L2 grammar are transferred wholesale to the L3. The paper further argues that a formal measure of the relative similarity of grammatical properties can established using the theory of the Contrastive Hierarchy (Dresher 2009). Arguments for this proposal come from studies of L3 English acquisition by L1 Arabic/L2 French speakers in Algeria and Tunisia. These studies suggest that some aspects of L3 English prosody, such as word stress, are determined largely on the basis of Arabic structures, whereas other aspects, such as rhythm, show a larger influence of French. With respect to segmental structure, similar patterns are found, with L3 English vowels being influenced by L2 French and L3 English consonants showing greater correspondence with L1 Arabic.
Evaluation:
The use of Contrastive Hierarchy theory in conjunction with Lightfoot’s parsing model provides a novel and insightful approach to the study of L3 acquisition. The examples used provide clear and convincing illustration of how these theoretical tools can be used to account for specific grammatical properties in the L3 and their relationship to L1 and L2 grammars. The paper also successfully contextualizes the proposals within big-picture debates in the field of L3 acquisition. Overall, I think the paper is well structured and convincingly argued and that it would make a significant contribution to the special issue at hand.
I found the most compelling aspect of the paper to be the discussion of the L3 English vowel system and its relationship to the feature hierarchies proposed for Arabic and French. The notions of complexity and parsing were clearly illustrated in this example. There were some details in the discussion that were confusing, however. The text states that English has twelve vowel phonemes but only eleven are illustrated in the hierarchies in (10) and (11). In relation to the hierarchy in (10), the text states that eight vowels are not uniquely specified but the list that follows contains nine vowels. It’s true that in the hierarchy in (10) none of the nine vowels listed are uniquely specified. But if we count the number of unique feature specifications given by the tree we get four unique sets. That would leave seven of the eleven vowel phonemes shown undifferentiated or eight of the twelve referred to earlier in the text.
Additional specific comments are noted below.
Comments:
pg. 2 – 3, Geopolitcal Context: The text states that Berber is the first language of 20-30% of the population in Algeria but also, based on data in Table 1, that Berber is used ‘very rarely’. This seemed odd. Is the population included in the data in Table 1 different from the general population in some way that would account for this? Also, the text uses the term Berber and the chart uses Tamazight. The chart would be easier to read if there was consistent use of the language name.
pg. 4, Figure 1 – The formatting of the trees is off here and throughout the document.
pg. 4 – 5, The description states that Benrabah’s data was transcribed by native English speakers. Are they phonetically trained transcribers? I was curious about how presence of pharyngealization was noted and whether this feature was salient to English listeners.
pg. 6, examples in (5) – A couple of sentences describing the structure of these examples would be helpful. Is this stress the same as L1 English, where are the relevant heavy syllables, etc.
pg. 7, line 281 – I’m a bit confused about what properties are referred to with respect to ‘property-by-property transfer’ here. I understood the preceding section to argue that the L3 English stress is based on the structure of Arabic stress in most cases, but that French stress patterns arise in some French/English cognates. If this is correct, I’m not sure what the different properties are in this example as both patterns relate to the property of word stress.
pg. 7, line 323 – The acronym PLD is used here but the term ‘primary linguistic data’ doesn’t appear in the paper until the next page.
pg. 9, line 421 – Acronym DMAP should be spelled out.
pg. 10 – 11, Explaining Consonants – This section makes use of feature geometry to model contrasts in laryngeal features. It’s not clear from the text to what extent the presented geometries are consistent with the theory of the Contrastive Hierarchy. A few sentences acknowledging the different representational assumptions in this section as compared to the preceding section would be helpful. Some of the representations used could be achieved using the Contrastive Hierarchy. For example, the difference between [spread glottis] languages and [voice] languages could follow from different ordering of these features in the hierarchies of the relevant languages. It might be useful to reference Hall (2003), which uses both feature geometry and Contrastive Hierarchy theory in an analysis of laryngeal features in West Slavic and may be compatible with the approach taken here. The relationship between the Contrastive Hierarchy and feature geometry is also discussed in Dresher (2009, chapter 5).
pg. 12 – 13, Levels of Comparison – I found the discussion of L2 English vowels as produced and perceived by L1 Mandarin speakers given here to be very useful in illustrating the points made in the preceding sections.
References – I didn’t do a thorough read-through of the references but noticed some items cited in the text are missing, such as Rothman (2015), Cowper and Hall (2019).
Comments for author File: Comments.pdf
Author Response
Please see attached
Author Response File: Author Response.pdf
