Skip to navigation – Site map

HomeIssues21VariaApplying Classification Trees to ...


Applying Classification Trees to Stress Patterns of Japanese Loanwords in English

Bradley Lunsford


This paper provides an analysis of the stress patterns of loanwords of Japanese origin in English. The ctree function of the party package in R is used to create classification trees. The stress pattern is not immediately obvious, so the algorithm is used to discover the significant predictor variables and create a classification tree. This illustration displays the stress pattern in a straightforward manner. The determining factor for the stress assignment of Japanese loanwords in English is syllable weight and vowel sonority. The presence of a final coda is the significant variable for the stress assignment in two-syllable loanwords, whereas vowel sonority is the significant variable for stress assignment in three and four-syllable loanwords. This provides support for Kenstowicz’s [1997] and deLacy’s [2002] universal sonority hierarchy.

Top of page

Full text

1. Introduction

1Loanwords are present in all world languages. “No language is entirely free from borrowed words, because no nation has ever been completely isolated” (Jespersen [1922: 208]). English, being an international language, is no exception. The English language has borrowed words from a variety of languages, mainly French and Latin. In a corpus study, Cannon, McArthur & Gachelin [1992] discovered that the English loanwords had origins from 84 different languages with French (25%), Spanish (8%), and Japanese (8%) making up the majority. Despite its geographic isolation, Japanese has exported many foreign words to other languages (Loveday [1996: 81]).

2There is a plethora of research on Japanese loanwords; however, this research primarily focuses on providing a purely descriptive listing of the loanwords (Abe [1990]; Cannon [1981], [1982]; Shogakukan [1994]; Sugiura [2003]). The purpose of this study is not simply to produce another descriptive listing of the loanwords, but to quantitatively investigate the stress patterns of Japanese loanwords. We will investigate whether or not the stress pattern can be predicated. For the purpose of this study, we will collect Japanese loanwords from the Oxford English Dictionary (OED) for analysis.

3In recent years, loanword phonology has become a productive research area providing much theoretical significance. However, suprasegmental adaptation from the borrowing language is understudied compared to segmental adaptation or phonotactic adaptation (Kang [2010], [2011]). In suprasegmental adaptation, choices are often made when there is no apparent motivation or there are more than one logical repair strategies. These phenomena of the emergent patterns in loanwords can pose a learnability puzzle (Broselow [2009]). This paper will investigate the emergent patterns of Japanese loanwords in English.

4The structure of the paper is as follows: Section 2 reviews general tendencies of the English stress pattern, as well as previous dictionary studies. In Section 3, the English stress pattern of Japanese loanwords is presented. We examine the possible predictor variables. Section 4 applies classification trees to the Japanese loanword data. In Section 5, we discuss the learnability of Japanese loanwords for English native speakers, and Section 6 concludes the paper.

2. Literature Review

2.1. English Stress Pattern

5Generally speaking, the most straightforward syllable stress rule for two syllable English words used by English language instructors is to stress the first syllable of nouns, adjectives, and adverbs. In contrast, stress the final syllable of verbs. This is seen below:

(1) ˈparent - ˈmember - ˈstudent [noun]
ˈquiet - ˈbasic - ˈbetter [
ˈmaybe - ˈnever - ˈseldom [
inˈclude - emˈbrace - neˈglect [

6In addition, there are a number of two syllable English words that are stressed on the first syllable when the word is being used as a noun, and stressed on the final syllable when being used as a verb, as seen in (2), so the stress assignment in English follows general tendencies (Roach [2009]).

(2) ˈproject [noun] / proˈject [verb]
ˈdesert [noun] / deˈsert [verb]
ˈprogress [noun] / proˈgress [verb]

7Three-syllable words are more complex due to the additional syllable. Most three syllable nouns, verbs, or adjectives are stressed on the first syllable, which is once again a general tendency. As a general rule, words terminating in “- er”, “-or”, “-ly”, or “-y” are stressed on the first syllable. Words terminating in “-tion”, “-sion”, “-ic”, or “al” generally have the stress on the penultimate syllable.

(3) ˈenergy - ˈfinally - ˈauditor
ˈcarrier - ˈcamera - ˈcharity
(4) conˈsumption - deˈnial - inˈvasio
hoˈrrific - caˈthedral - tranˈsaction

8Lastly, words terminating in “-ee”, “-eer”, “ese”, “ette”, or “-ique” generally have the stress on the final syllable. On the other hand, four-syllable English words are stressed on the antepenultimate or penultimate syllable. Generally, words terminating in “-cy”, “-ty”, “-phy”, “- gy”, or “al” receive antepenultimate stress. To sum up, like three-syllable words, four-syllable words terminating in “-tion”, “-sion”, or “-ic” receive penultimate stress (Roach [2009]).

(5) engiˈneer - volunˈteer - refeˈree
Japaˈnese - silhouˈette - Mozamˈbique
(6) eˈquality - psyˈchiatry - deˈpendency
phoˈtography - psyˈchology - iˈllogical
(7) unreaˈlistic - teleˈvision - abdiˈcation

9These stress rules are often memorized by second language learners on their quest to improve their English pronunciation. It needs to be emphasized that these are just some general tendencies of the English language. Next, we will examine some previous dictionary studies.

2.2. Previous Dictionary Studies

10Before embarking on our journey, let us first look at previous dictionary studies on loanword phonology. The frequency distribution of English stress patterns has been previously investigated in Clopper [2002]. Clopper [2002] calculated the word frequencies based on the data provided by the Hoosier Mental Lexicon which is an online version of Webster’s Pocket Dictionary. Figure 1 below shows the breakdown of the multisyllabic words. The table shows the position of stress for English words containing 2, 3, and 4 syllables respectively.

Figure 1. Position of Primary Stress in English

Figure 1. Position of Primary Stress in English

11We can conclude from the table that two and three-syllable words are the most frequent in the English language and there is a tendency for these words to exhibit stress on the initial syllable. The most common stress position for two-syllable words is penultimate, while the most common stress position for three-syllable words is antepenultimate. Four syllable words are mostly stressed on the antepenultimate syllable. Overall, two-syllable words are the most common in the English language.

12To date, many phonological studies have been conducted regarding English loanwords in other languages; however, few studies have been conducted the other way around. Fournier [2018] discovered Italian loanwords in English preserve their original Italian stress patterns in <VCV#> structures. This means that English faithfully reproduces the phonological properties of Italian stress. Fournier investigated loanwords of Italian origin in the OED and discovered that out of the 309 loanwords, 296 words are total stress matches no matter what the original Italian stress pattern may be. Final, penultimate, and antepenultimate stress patterns in Italian are all preserved in English.

2.3. Suprasegmental Adaptation

13The position of pitch prominence (variations of high tone) is relatively unrestricted in tonal languages; thus, these languages do not prevent the faithful preservation of input language prominence (Kang [2010]). The stress of the input language is preserved as a high tone variant in a tonal language.

(8a) English to Cantonese (Yip [2006])
mótor >> mɔH taMH

(8b) English to Yoruba (Kenstowicz [2006])
tomáto >> toM
maH toL

(8c) English to Hausa (Kenstowicz [2006])
recéipt >> raL

(8d) English to Shona (Kenstowicz [2006])
récipe >>
reH siL piL

(8e) English to Twi (Devonish ([2002])
búcket >>
buH kiL tiL

(8f) Portuguese to Kongo (Devonish [2002])
rúa >>
lu:LH laL ‘street’

14Pitch accent languages usually behave differently from tonal languages. Shinohara [2004] shows that the pitch of loanwords of French origin in Japanese is assigned by the Latin stress rule. That means the accent is placed on the head of a bimoraic trochee at the right edge with final syllable extrametricality.

(9a) French loanwords in Japanese
travɛsti >> torabésuti ‘travesty’

(9b) English loanwords in Japanese
techníque >> tekuní:ku

15On the other hand, the stress of loanwords of English origin is preserved for the most part (Shinohara [2000], [2004]).

3. The phenomenon: English stress patterns of Japanese loanwords

3.1. Two-Syllable Japanese Loanwords

16This paper will investigate the predictability of stress patterns in Japanese loanwords. 208 two-syllable Japanese loanwords were collected from the OED. The complete list of Japanese loanwords is available in the appendix. I will use the OED data to investigate the possible predictability of the choice between initial stress and final stress. The two stress variants are illustrated below.

(10a) fúton - háiku - rámen - súmo
(10b) tycóon - Kirín - Sendái - typhóon

In (10a), the stress is placed on the initial syllable, while in (10b) stress is placed on the final syllable. We will address the question regarding which factors motivate the choice between the initial stress and final stress in the next section. The possible predictor variables are consonant quality (obstruent/sonorant), vowel quality (low/non-low), initial/final diphthong, and initial/final coda. Initial/final coda refers to the placement of consonants in relation to vowels within the initial or final syllable.
Initial Onset Quality: whether the initial onset of the source word is an obstruent or sonorant.
Values: obstruent / sonorant

(11a) futon - sumo - tycoon - typhoon

(11b) manga - miso - Nippon - Meiji

The initial onset in (11a) is an obstruent, while the initial onset in (11b) is a sonorant.
Final Onset Quality: whether the final onset of the source word is an obstruent or sonorant. Values: obstruent / sonorant

(12a) sushi - futon - tycoon - typhoon
(12b) sumo - ramen - kana - sumi

The final onset in (12a) is an obstruent and the final onset in (12b) is a sonorant.
Initial Vowel Quality: whether the initial vowel is a low vowel or non-low vowel. Values: low / non-low

(13a) ramen - sake - baka - kana
(13b) sushi - sensei - judo - sumo

The initial vowel in (13a) is a low vowel, while the initial vowel in (13b) is a non-low vowel.
Final Vowel Quality: whether the final vowel is a low vowel or non-low vowel. Values: low / non-low

(14a) kata - futon - Nippon - tsuba
(14b) guru - ninja - haiku - judo

The final vowel in (14a) is a low vowel. On the contrary, the final vowel in (14b) is a non-low vowel.
Initial/Final Diphthong: whether the initial or final syllable contains a diphthong.

(15a) geisha - haiku - gaijin - kaiken
(15b) banzai - bonsai - sensei - Sendai

The initial syllable in (15a) contains a diphthong, while the final syllable in (15b) contains a diphthong.
Initial/Final Coda: whether the initial or final syllable contains a coda.

(16a) bento - bonsai - ninja - gingko
(16b) futon - ramen - tycoon - Kirin

The initial syllable in (16a) contains a coda and the final syllable in (16b) contains a coda.

17This subsection discusses the possible predictor values that we will put into our classification tree. At this point, we are not sure which factors are crucial for determining whether a two-syllable Japanese loanword receives initial stress or final stress. Looking ahead, whether or not the loanword contains a coda in the final syllable is a key factor for determining stress assignment; however, at this point, we do not know this yet.

3.2 Three-Syllable Japanese Loanwords

18In addition to two-syllable Japanese loanwords, this paper will also investigate the predictability of the stress patterns in three-syllable Japanese loanwords. 160 three-syllable Japanese loanwords were collected from the OED. These loanwords are also located in the appendix of this paper.

19The stress variants in three-syllable Japanese loanwords are illustrated by the following examples.

(17a) emóji - sashími - kimóno - sudóku
(17b) ánime - sámurai - Súbaru - híbachi

For three-syllable Japanese loanwords, the stress can either fall on the penultimate syllable as seen in (17a) or the antepenultimate syllable (17b). The possible predictor variables are onset quality (obstruent, sonorant, no onset) and vowel quality (low, non-low). Looking at the examples above:

20/emoji/ is comprised of the following predictor variables:

  • Initial Onset: no onset Initial Vowel: non-low

  • Middle Onset: sonorant Middle Vowel: non-low

  • Final Onset: obstruent Final Vowel: non-low

21Likewise, /anime/ is comprised of the following predictor variables:

  • Initial Onset: no onset Initial Vowel: low

  • Middle Onset: sonorant Middle Vowel: non-low

  • Final Onset: sonorant Final Vowel: non-low

22The possible predictor variables mentioned above will be entered into our classification tree to determine which ones predict stress assignment to a high degree of certainty. We will be specifically looking at consonant quality and vowel quality.

3.3. Four-Syllable Japanese Loanwords

23Lastly, we will investigate the predictability of the stress pattern in four-syllable Japanese loanwords. 42 Japanese loanwords were collected from the OED. Like the previous loanwords, these loanwords can be found in the appendix of this paper. The stress variants in four-syllable Japanese loanwords are illustrated by the following examples.

(18a) Hiróshima - guríbori - nikúbori - karáteka
(18b) yokozúna - Ikebána - hiragána - kirigámi

(18a) exhibits antepenultimate stress and (18b) exhibits penultimate stress. For the predictor variables, we will consider onset quality (obstruent, sonorant, no onset) and vowel quality (low, non-low).

24In /hiroshima/, we observe the following predictor variables:

  • Initial Onset: obstruent Initial Vowel: non-low

  • Middle Onset1: sonorant Middle Vowel1: non-low

  • Middle Onset2: obstruent Middle Vowel2: non-low

  • Final Onset: sonorant Final Vowel: non-low

25In addition, /yokozuna/ contains the following predictor variables:

  • Initial Onset: sonorant Initial Vowel: non-low

  • Middle Onset1: obstruent Middle Vowel1: non-low

  • Middle Onset2: obstruent Middle Vowel2: non-low

  • Final Onset: sonorant Final Vowel: non-low

26We will also investigate four-syllable Japanese loanwords based on consonant quality and vowel quality. It will be interesting to see whether four-syllable Japanese loanwords pattern with three-syllable Japanese loanwords, or if they pattern independently.

27In the following section, we will demonstrate the application of classification trees to the phonological data. Since there is a complex interaction among the variables, classification trees are a suitable choice.

4. Classification Trees

28Classification trees (Hothorn, Hornik & Zeileis [2006]) utilize a recursive partitioning algorithm which divides the sample into subsamples based on the independent variables. This algorithm tests whether the predictor variables are significantly associated with the dependent variable. In our case, the dependent variable is the location of the stress. The subsamples are displayed in a tree diagram with the variable exhibiting the strongest association at the top.

29The algorithm will go through the following steps: (1) Test the global null hypothesis of independence between any of the input variables and the response. Stop if the hypothesis cannot be rejected or select the input variable with the strongest association to the response. The association is measured by a p-value corresponding to the test for the partial null hypothesis of an input variable and the response. (2) Implement a binary split in the input variable. (3) Recursively repeat steps (1) and (2). For our analysis, the minimum criterion is set at 0.95 which means that the p-value must be smaller than 0.05 in order to spit the node. This approach ensures that the right size tree is grown without the need for additional pruning.

30In the following subsections, we will create classification trees with the ctree function of the party package in R (R Core Team [2022]). After we downloaded the party package in R, the csv files of the phonological data were uploaded for analysis.

4.1. Two Syllable Japanese Loanwords

31We calculated a classification tree using the ctree function to predict the location of stress (initial stress: yes/no) based on the predictor variables, initial onset, initial vowel, initial diphthong, initial coda, final onset, final vowel, final diphthong, and final coda. The application resulted in the following tree model.

Figure 2. Two-Syllable Japanese Loanwords

Figure 2. Two-Syllable Japanese Loanwords

32The classification tree produces binary splits of the data, thereby creating subsamples of the data. The present tree features two splits, resulting in five nodes. The first split (node 1) is created by the variable ‘FinCoda’ (final syllable coda: yes/no). It shows that the probability is high that two syllable Japanese loanwords with initial stress will not contain a coda in the final syllable (see node 3). Also, two syllable Japanese loanwords with final stress will likely contain a coda in the final syllable. Obtaining a p-value less than 0.05 shows that the data is statistically significant. Overall, initial stress is preferred in two syllable Japanese loanwords.

33The interaction between the variables ‘FinCoda’ and ‘FinV” is the only significant influence on the stress choice for two-syllable Japanese loanwords. The other predictor variables are found to have no significant influence. Adding the other variables to the tree would not increase its prediction power. If there is a diphthong and a non-low vowel in the final syllable, the model can predict the stress placement with nearly 100% accuracy.

34The list of two-syllable Japanese loanwords that do not follow the pattern of initial stress are as follows:

(19a) fuchí
(19b) gobán
(19c) Hizén
(19d) zazén
(19e) kaizén
(19f) Kirín
(19g) mokkó
(19h) typhóon
(19i) Sendái
(19j) seppá
(19k) tokkín
(19l) tycóon

35Our findings that a syllable final coda is the primary determining factor for assigning stress in two-syllable Japanese loanwords is on par with Ross’s [1972] findings that there are stressing tendencies governed by the final consonant of a word in English. These are only tendencies, as seen above. For example, (19j) does not contain a final syllable coda.

36We also determined the presence of a final syllable diphthong to be a determining factor for stress assignment. According to Chomsky & Halle [1968], stress placement is sensitive to syllable weight. By definition, a weak syllable cluster is “a string consisting of a simple vocalic nucleus followed by no more than one consonant,” and a strong syllable cluster is “a string consisting of either a vocalic nucleus followed by two or more consonants or a complex vocalic nucleus followed by any number of consonants” (Chomsky & Halle [1968:  29]). Following this train of thought, codas in the final syllable of two-syllable Japanese loanwords would impact the placement of stress because it affects the syllable weight. The presence of final syllable codas in our classification tree is no surprise. In these loanwords, syllables with a coda are heavy syllables, while syllables without a coda are light. This can be demonstrated below:

Table 1. zazen

Table 1. zazen

37Look at the leftmost column of the table. This column lists the candidate forms. These are the different possible pronunciations of a word or phrase based on the constraints. Across the top of the table, you will see a row of constraints. These constraints represent linguistic principles that guide the evaluation of the candidates. They are often abbreviated with symbols or shorthand, which is explained below. In the cells of the tableau, you will see an asterisk if a constraint has been violated by a candidate. The exclamation point indicates that the candidate has been eliminated. The top-ranked constraints are more important and have a higher priority. The candidate with the fewest violations is the winner. This candidate represents the optimal pronunciation according to the constraint ranking.

38IO-OI: Output segments must have input correspondents, and input segments must have output correspondents
Parse-syllable: A syllable must be footed
Weight-to-Stress Principle (WSP): Heavy syllables must be stressed

39The conflict among the constraints in the table above must be resolved in order to assign stress. The first candidate does not violate WSP making it the optimal candidate receiving final stress. The other three candidates receive fatal violations leading to improper stress placement.

4.2. Three-Syllable Japanese Loanwords

40Next, we calculated a classification tree for three-syllable Japanese loanwords. The classification tree is used to predict the stress location (penultimate stress: yes/no). If the stress is not located on the penultimate syllable, it is located on the antepenultimate syllable. Based on the predictor variables, the ctree function generated the classification tree below. The first split is created by the variable ‘MidV’ (1: low, 0: non-low). Three-syllable Japanese loanwords containing a low middle vowel have a high probability to exhibit penultimate stress. The interaction between the predictor variables ‘MidV’ and ‘FinV’ (1: low, 0: non-low) is the significant factor determining whether the three-syllable Japanese loanword will contain penultimate stress or antepenultimate stress. Three-syllable Japanese loanwords tend to prefer penultimate stress; however, it is worth noting that loanwords containing a non-low middle vowel and a low final vowel demonstrate a preference for antepenultimate stress.

41The p-values of p < 0.001 and p = 0.01 show that our data is statistically significant. The predictability of three-syllable loanwords utilizes sonority as the determining factor, as opposed to two-syllable loanwords, which rely on the presence of a coda or a diphthong. If the loanword contains a low vowel in the middle syllable, we can predict penultimate stress with nearly 100% accuracy.

42Our findings support a sonority-based stress assignment system for three-syllable Japanese loanwords. According to Kenstowicz [1997], a “sonority-driven” stress system is one where the relative sonority of the syllabic nuclei helps determine the optimal stress bearing unit. The universal sonority hierarchy is as follows (Kenstowicz [1997]; de Lacy [2002]):

(20) low peripheral > mid peripheral > high peripheral > mid peripheral > high peripheral

43The overwhelming tendency is for words possessing penultimate stress to contain low vowels in the penultimate syllable, examples in (21).

(21) tsunámi - wasabi - tatámi - karáte

Figure 3. Three-Syllable Japanese Loanwords

Figure 3. Three-Syllable Japanese Loanwords

44The two loanwords that do not follow the tendency are Shinkansen and wakame. Shinkansen and wakame both contain a low vowel in the penultimate syllable; however, they demonstrate antepenultimate stress. All other three syllable Japanese loanwords containing low vowels in the penultimate syllable exhibit penultimate stress. This data presents strong evidence for Kenstowicz’s sonority hierarchy.

45Below is an Optimality Theory (OT) analysis for three-syllable Japanese loanwords. Table 2 shows the optimal candidate for the loanword karate, and Table 3 shows the optimal candidate for the loanword Subaru.

Table 2. karate



*HDFT/ i,u


☞ kə(‘rati)




Table 3. subaru



*HDFT/ i,u




☞ (‘subə)ru



46ALIGN-R (Ft,PrWd): The right edge of every foot must coincide with the right edge of a PrWd
*HDFT: Incur a violation for every head that contains a mid-central vowel
*HDFT/ i,u: Incur a violation for every head that contains a high peripheral vowel

4.3. Four-Syllable Japanese Loanwords

47Lastly, we used the ctree function to generate the classification tree for four-syllable Japanese loanwords. According to the results of the algorithm, the significant predictor variable is ‘Mid2V’ (1: low, 0: non-low). Mid2V represents the vowel in the penultimate syllable. The classification tree is displayed below in figure 4. The classification tree contains two splits and five nodes. If the loanword contains a low vowel in the penultimate syllable, then it almost always exhibits penultimate stress. A p-value of 0.031 is reliable; however, the IniV (p = 0.622) is not very reliable and most likely not a crucial factor for determining stress placement. Below is a list of the loanwords demonstrating antepenultimate stress. Wakizashi is the only loanword which does not conform to the tendency. Here a low vowel is present, but the loanword receives antepenultimate stress.

(22) Kakíemon - guríbori - karáteka
nikúbori - wakízashi - Hiróshima

Figure 4. Four-Syllable Japanese Loanwords

Figure 4. Four-Syllable Japanese Loanwords

48In this section, we have created classification trees and analyzed our data sets. Once again, the loanwords conform to Kenstowicz’s sonority hierarchy. However, with only 42 tokens, the statistical power of our four-syllable data set is not as robust as the other two data sets.

49In Japanese, the simplicity and consistency of the vowel system contribute to the clarity and distinctiveness of Japanese pronunciation. Each vowel sound is distinct and does not vary significantly in different contexts, making Japanese phonetics relatively straightforward compared to languages with more complex vowel systems. Our analysis investigated the low vowel /a/, mid vowels /e/ and /o/, and the high vowels /i/ and /u/. A significant factor for determining stress placement for three or four-syllable Japanese loanwords is whether or not the vowel is low or non-low.

4.4. Discussion

50According to Breiman et al. [1984], classification trees are capable of capturing complex decision boundaries and interactions among variables, which can lead to accurate predictions. Additionally, they provide interpretability, as the resulting tree structure can be easily visualized and understood by domain experts. However, it is important to note that classification trees are prone to overfitting, especially when the tree becomes too deep and complex. This can result in poor generalization to new data. Efforts to mitigate overfitting include techniques like pruning, limiting tree depth, and using ensemble methods like Random Forests (Breiman [2001]). The data in this study generated classification trees with only five nodes, so the trees are relatively simple and easy to interpret.

51This study uses a classification tree over a random forest, because interpretability is key. We prioritize understanding the decision making process and want a clear, visual representation of the model’s decisions, so a single classification tree can provide a more straightforward explanation compared to the ensemble nature of a random forest. Also, these classification trees serve as a starting point for exploring the data and gaining insights into the relationships between features and the target variable. Now that we have a better understanding of the data, we can decide to use a more complex model like random forests in the future. Lastly, training a random forest involves constructing multiple decision trees, which requires more computational resources and time compared to training a single decision tree. For this initial investigation, a single decision tree was more feasible. In the future, a random forest could be utilized over a single decision tree, as random forests typically offer better predictive accuracy and robustness compared to a single classification tree.

5. Learnability

52Lastly, we will discuss the learnability of Japanese loanwords in English from the point of view of English native speakers. It has been proposed that the phonological acquisition model involves a set of constraints where markedness constraints are ranked above faithfulness constraints at the initial state (Gnanadesikan [2004]). This ensures that the acquisition begins with the most restrictive grammar. Markedness constraints tend to be more restrictive than faithfulness constraints in phonology because they reflect language-specific preferences for simplicity and economy in linguistic structures. Faithfulness constraints prioritize preserving the original input or underlying form. They strive to maintain similarity between the input and output. Faithfulness constraints allow for exceptional cases by permitting deviations from marked or regular patterns when the input itself is marked or irregular.

(23) Possible Ranking Patterns in Loan Adaptation (Brosleow [2009])
markedness > faithfulness (default pattern)
faithfulness > markedness (motivated by marked forms)
markedness > markedness, faithfulness > faithfulness (mysterious if not data driven)

53Japanese does not contain stress in the same sense as English; however, words can contain a pitch accent. In Japanese, a difference in pitch accent can be phonemic and produce minimal pairs.

(24a) háshi - chopsticks (first syllable accent)
(24b) hashí - bridge (second syllable accent)
(24c) hashi - edge (accentless)

54Even though pitch accent can be phonemic in Japanese, approximately 70% of native nouns are unaccented (Labrune [2012]). English speakers acquiring Japanese loanwords must employ different repair strategies because the two stress structures are equally impossible in the borrowing language. Japanese does not have a stress accent like English. This made the acquisition pattern not apparently obvious at first glance. However, our classification trees bought to light an interesting phenomenon where final syllable codas and vowel sonority are crucial predictor variables for Japanese loanwords in English.

6. Conclusion

55Classification trees have recently become a new methodological tool in the field of quantitative linguistics (Gries [2019]; Ni et al. [2010]). Classification trees are chosen when predictor variables are involved in complex interactions (James et al. [2013]). The results of a logistic regression analysis are hard to interpret with complex interactions and trees offer a more intuitive illustration of the results (Tagliamonte & Baayen [2012]).

56We successfully applied classification trees to the present case study of Japanese loanwords in English. This new methodological tool can handle complex interactions among predictor variables and present a straightforward illustration of the interaction effects. The illustrations provide an easy interpretation of the possible interactions among predictor variables for Japanese loanwords.

57Shih & deLacy [2019] argue that the vast majority of evidence comes from “impressionistic descriptions” which are considered to be “any description that involves a human perceiving and reporting the position of stress, unaided by mechanical analytical devices.” The R software provided us with a mechanical device to perform a computational analysis mitigating the need for “impressionistic descriptions.” The classification trees present evidence for a sonority-driven stress system.

Top of page


Abe H., 1990, “Gaikoku ni shinshutsu shita nihongo tachi” [Japanese words that made their ways into the world], Katakanago no Joshiki Hijoshiki [Common sense and absurdity about katakana words]; Tokyo: Tokyo Soseki, 101-144.

Breiman L., Friedman J. H., Olshen R. A. & Stone C. J., 1984, Classification and regression trees, Wadsworth & Brooks.

Breiman L., 2001, “Random forests”, Machine Learning 45(1), 5-32.

Brosleow E., 2009, “Stress adaptation in loanword phonology: perception and learnability”, in Boersma P. & Hamann S. (Eds.), Phonology in perception, Berlin: Mouton de Gruyter, 11-58.

Cannon G., 1981, “Japanese borrowings in English”, American Speech 56, 190-206.

Cannon G., 1982, “698 Japanese loanwords in English”, Verbatim 9, 9-10.

Cannon G., McArthur T. & Gachelin J., 1992, “Borrowing”, in McArthur Tom (Ed.), The Oxford Companion to the English Language, Oxford: Oxford University Press, 141-145.

Chomsky N. & Halle M., 1968, The Sound Pattern of English, New York: Harper and Row (reprinted in 1991 by MIT Press).

Clopper S., 2002, “Frequency of stress patterns in English: A computational analysis”, Indiana University Linguistics Club Working Papers 2.

de Lacy P., 2002, The formal expression of markedness, PhD dissertation, Amherst: University of Massachusetts.

Devonish H., 2002, Talking Rhythm, Stressing Tone: The Role of Prominence in Anglo-West-African Creole Languages, Kingston: Arawak Publications.

Fournier P., 2018, “Stress Assignment in Italian Loanwords in English and its Impact on the Stressing of Foreign Words by Native English Speakers”, Corela [Online], HS-24.

Gnanadesikan A., 2004, “Markedness and faithfulness constraints in child phonology”, Constraints in Phonological Acquisition, Cambridge University Press.

Gries S., 2019, “On classification trees and random forests in corpus linguistics: Some words of caution and suggestions for improvement”, Corpus Linguistics and Linguistic Theory 16(3), 617-647.

Hothorn T., Hornik K. & Zeileis A., 2006, “Unbiased Recursive Partitioning: A Conditional Inference Framework”, Journal of Computational and Graphical Statistics 15(3), 651-674.

James G., Witten D., Hastie T. & Tibshirani R., 2013, An introduction to statistical learning, Vol. 112, New York: Springer, 18.

Jespersen O., 1922, Language: Its Nature, Development and Origin, London: George Allen and Unwin.

Kang Y., 2010, “Tutorial overview: Suprasegmental adaptation in loanwords”, Lingua 120, 2295-2310.

Kang Y., 2011, “Loanword phonology”, The Blackwell companion to phonology, 1-25.

Kenstowicz M., 1997, “Quality-sensitive stress”, Rivista di Linguistica 9, 157-188.

Kenstowicz M., 2006, “Tone loans: the adaptation of English loanwords into Yoruba”, in Mugabe J. (Ed.), Proceedings of Selected Proceedings of the 35th Annual Conference on African Linguistics, Cascadilla Proceedings Project, 136–146.

Labrune L., 2012, The phonology of Japanese, Oxford: Oxford University Press, 186-188.

Loveday L. J., 1996, Language Contact in Japan: A Socio-Linguistic History, New York: Oxford University Press.

Ni C. J., Liu W. J. & Xu B, 2010, “Mandarin stress detection using hierarchical model based boosting classification and regression trees”, The 2010 International Joint Conference on Neural Networks (IJCNN), IEEE, 1-5.

R Core Team, 2022, R: A language and environment for statistical computing, R Foundation for Statistical Computing, Vienna, Austria:

Roach P., 2009, English Phonetics and Phonology, Cambridge University Press.

Ross J. R., 1972, “A Reanalysis of English Word Stress”, in Brame M. (Ed.), Contributions to Generative Phonology, Austin and London: University of Texas Press.

Shogakukan (Ed.), 1994, “Nihongo kara shakuyo-sareta eigo” [English words borrowed from Japanese], Shogakukan Random House English-Japanese Dictionary, 2nd ed., Tokyo: Shogakukan, 3181-3185.

Shih S.-H. & deLacy P., 2019, “Evidence for Sonority-Driven Stress”, Catalan Journal of Linguistics 18, 9-40.

Shinohara S., 2000, “Default accentuation and foot structure in Japanese: evidence from Japanese adaptations of French words”, Journal of East Asian Linguistics 9, 55–96.

Shinohara S., 2004, “Emergence of universal grammar in foreign word adaptations”, in Kager R., Pater J. & Zonneveld W. (Eds.), Constrains in Phonological Acquisition, Cambridge: Cambridge University Press, 292–320.

Sugiura M., 2003, “Eigo no naka no nihongo shakuyogo” [Japanese loanwords in English], in Harunori Hisada (Ed.), Bunka no Kareidosukopu: Eibei Gengo Bunka Ronshu [The kaleidoscope of culture: Collection of essays on English and American language and culture], Tokyo: Eihosha, 215-224.

Tagliamonte S. A. & Baayen R. H., 2012, “Models, forests and trees of York English: Was/were variation as a case for statistical practice”, Language Variation and Change 34, 135-178.

Yip M., 2006, “The symbiosis between perception and grammar in loanword phonology”, Lingua 116, 950-975.

Top of page


List 1: Two Syllable Japanese Loanwords

(1) Baka, (2) Bancha, (3) Banzai, (4) Baren, (5) Basho, (6) Beddo, (7) Bekko, (8) Bento, (9)

Beta, (10) Biwa, (11) Bizen, (12) Bonsai, (13) Budo, (14) Butoh, (15) Daikon, (16) Daimyo,

(17) Dairi, (18) Daisho, (19) Dango, (20) Dashi, (21) Demo, (22) Dojo, (23) Fuchi, (24) Fugu,

(25) Fuji, (26) Furo, (27) Futon, (28) Gaijin, (29) Geisha, (30) Gengo, (31) Genro, (32) Geta,

(33) Gingko, (34) Goban, (35) Gobo, (36) Guru, (37) Gyoza, (38) Habu, (39) Hagi, (40) Haiku,

(41) Happi, (42) Heimin, (43) Heisei, (44) Hiba, (45) Hinin, (46) Hizen, (47) Honcho, (48)

Hoochie, (49) Janken, (50) Jito, (51) Jodo, (52) Jomon, (53) Joro, (54) Judo, (55) Juku, (56)

Junshi, (57) Kago, (58) Kaiken, (59) Kainic, (60) Kaizen, (61) Kaki, (62) Kakke, (63) Kami, (64)

Kana, (65) Kanban, (66) Kanji, (67) Kanten, (68) Kata, (69) Katsu, (70) Kaya, (71) Kegon, (72)

Kembei, (73) Kendo, (74) Kiku, (75) Kikyo, (76) Kimon, (77) Kirin, (78) Koban, (79) Kobang,

(80) Kobe, (81) Kogai, (82) Koicha, (83) Koji, (84) Koku, (85) Kombu, (86) Kondo, (87) Koro,

(88) Koto, (89) Koza, (90) Kozo, (91) Kudzu, (92) Kuge, (93) Kura, (94) Kuzu, (95) Kylin, (96)

Kyudo, (97) Machi, (98) Maiko, (99) Mako, (100) Mana, (101) Manga, (102) Maru, (103)

Matsu, (104) Mebos, (105) Meiji, (106) Mikan, (107) Mingei, (108) Mirin, (109) Miso, (110)

Mokko, (111) Mokume, (112) Mome, (113) Mompei, (114) Mondo, (115) Moxa, (116) Mume,

(117) Mura, (118) Nanga, (119) Napa, (120) Nara, (121) Nashi, (122) Nikkei, (123) Ninja, (124)

Nippon, (125) Nisei, (126) Nori, (127) Noshi, (128) Notan, (129) Raku, (130) Ramen, (131)

Reiki, (132) Renga, (133) RichShaw, (134) Rikka, (135) Ritsu, (136) Roju, (137) Ronin, (138)

Roshi, (139) RyuKyu, (140) Sabi, (141) Sake, (142) Sansei, (143) Seiza, (144) Sendai, (145)

Sennin, (146) Senryu, (147) Sensei, (148) Seppa, (149) Seto, (150) Shaku, (151) Shiga, (152)

Shijo, (153) Shimpa, (154) Shingon, (155) Shinto, (156) Shishi, (157) Shochu, (158) Shogi,

(159) Shogun, (160) Shoji, (161) Shokku, (162) Showa, (163) Shoyu, (164) Shugo, (165)

Shunga, (166) Shunto, (167) Sika, (168) Skimmia, (169) Soba, (170) Soshi, (171) Soto, (172)

Sugi, (173) Sumi, (174) Sumo, (175) Sushi, (176) Tabi, (177) Tachi, (178) Taisho, (179) Tamo,

(180) Tanka, (181) Tansu, (182) Tanto, (183) Tempo, (184) Tendai, (185) Tenko, (186) Tofu,

(187) Tojo, (188) Tokkin, (189) Torii, (190) Tosa, (191) Tsuba, (192) Tsubo, (193) Tycoon,

(194) Typhoon, (195) Wabi, (196) Wagyu, (197) Waka, (198) Washi, (199) Yagi, (200) Yayoi,

(201) Yedda, (202) Yugen, (203) Yusho, (204) Yuzen, (205) Zaikai, (206) Zazen, (207) Zendo,

(208) Zori

List 2: Three Syllable Japanese Loanwords

(1) Adsuki, (2) Aikido, (3) Aikuchi, (4) Akita, (5) Amado, (6) Anime, (7) Arita, (8) Aucuba, (9)

Awabi, (10) Betamax, (11) Bugaku, (12) Bunraku, (13) Bushidao, (14) Butsudan, (15) Byobu,

(16) Chanoyu, (17) Daibutsu, (18) Daruma, (19) Dokusan, (20) Emoji, (21) Endaka, (22)

Engawa, (23) Enoki, (24) Funori, (25) Fusuma, (26) Gagaku, (27) Genroku, (28) Godzilla, (29)

Gomoku, (30) Habatsu, (31) Habutai, (32) Hakama, (33) Hanami, (34) Haniwa, (35) Haori, (36)

Haragei, (37) Hayashi, (38) Hechima, (39) Hibachi, (40) Hijiki, (41) Hinoki, (42) Hirado, (43)

Iaido, (44) Ibota, (45) Ichibu, (46) Imari, (47) Iroha, (48) Ishime, (49) Jinricksha, (50) Joruri,

(51) Judogi, (52) Judoka, (53) Jujitsu, (54) Kabane, (55) Kabocha, (56) Kabuki, (57) Kagura,

(58) Karate, (59) Karatsu, (60) Karoshi, (61) Kashira, (62) Katana, (63) Katsura, (64) Kazoku,

(65) Keiretsu, (66) Kenjutsu, (67) Keyaki, (68) Kikumon, (69) Kimono, (70) Kirimon, (71)

Kokeshi, (72) Kotatsu, (73) Kumaso, (74) Kumite, (75) Kuruma, (76) Kurume, (77) Kutani, (78)

Kuzushi, (79) Kyoto, (80) Mama-san, (81) Mamushi, (82) Marumi, (83) Matsuri, (84) Medake,

(85) Menuki, (86) Mikado, (87) Mishimi, (88) Muraji, (89) Nagami, (90) Nakodo, (91) Nanako,

(92) Nandina, (93) Narikin, (94) Netsuke, (95) Ninjutsu, (96) Nintendo, (97) Nogaku, (98)

Norimon, (99) Norito, (100) Nunchaku, (101) Odori, (102) Ojime, (103) Otaku, (104) Pachinko,

(105) Randori, (106) Romaji, (107) Rotenone, (108) Rumaki, (109) Ryokan, (110) Ryuku, (111)

Sakaki, (112) Sakura, (113) Samisen, (114) Samurai, (115) Senpaku, (116) Sasanqua, (117)

Sashimi, (118) Satori, (119) Sastuma, (120) Sawara, (121) Sentoku, (122) Seppuku, (123)

Shakudo, (124) Shiatsu, (125) Shibui, (126) Shigella, (127) Shiitake, (128) Shikimi, (129)

Shimada, (130) Shingeki, (131) Shinkansen, (132) Shizoku, (133) Shotokan, (134) Shubunkin,

(135) Shuriken, (136) Sobaya, (137) Sodoku, (138) Sokaiya, (139) Subaru, (140) Sudoku, (141)

Surimi, (142) Suzuki, (143) Tamari, (144) Tatami, (145) Temmoku, (146) Tempura, (147)

Tsunami, (148) Uchiwa, (149) Ukemi, (150) Urushi, (151) Wakame, (152) Warabi, (153)

Wasabi, (154) Yakuza, (155) Yamato, (156) Yashiki, (157) Yukata, (158) Yukawa, (159)

Zabuton, (160) Zaibatsu

List 3: Four Syllable Japanese Loanwords

(1) Guribori, (2) Hanamichi, (3) Hanashiki, (4) Harakiri, (5) Hashimoto, (6) Hatamoto, (7)

Hibakusha, (8) Hiragana, (9) Hiroshima, (10) Ikebana, (11) Ishihara, (12) Kakemono, (13)

Kakiemon, (14) Kamikaze, (15) Karaoke, (16) Karateka, (17) Katakana, (18) Katayama, (19)

Kawasaki, (20) Kirigami, (21) Kurikata (22) Kuromaku, (23) Makimono, (24) Nikubori, (25)

Okinawa, (26) Origami, (27) Sayonara, (28) Shabu-Shabu, (29) Shakuhachi, (30) Soka Gakkai,

(31) Sukiyaki, (32) Surimono, (33) Teppanyaki, (34) Teriyaki, (35) Tokonoma, (36) Tokugawa,

(37) Tsukemono, (38) Wakizashi, (39) Yakitori, (40) Yokohama, (41) Yokozuna, (42) Yoshiwara

Top of page

List of illustrations

Title Figure 1. Position of Primary Stress in English
File image/png, 73k
Title Figure 2. Two-Syllable Japanese Loanwords
File image/png, 5.8k
Title Table 1. zazen
File image/png, 5.3k
Title Figure 3. Three-Syllable Japanese Loanwords
File image/png, 66k
Title Figure 4. Four-Syllable Japanese Loanwords
File image/png, 9.8k
Top of page


Electronic reference

Bradley Lunsford, Applying Classification Trees to Stress Patterns of Japanese Loanwords in EnglishLexis [Online], 21 | 2023, Online since 13 November 2023, connection on 02 December 2023. URL:; DOI:

Top of page

About the author

Bradley Lunsford

National Tsing Hua University, Taiwan

Top of page



The text only may be used under licence CC BY-SA 4.0. All other elements (illustrations, imported files) are “All rights reserved”, unless otherwise stated.

Top of page
Search OpenEdition Search

You will be redirected to OpenEdition Search