A corpus study of phonological factors in novel English blends⇤

Michael Dow
Université de Montréal

1. Introduction

A commonly cited limitation on the study of peripheral, extragrammatical and/or ludic pro-
cesses in morphophonology is the relative difficulty of automated retrieval of their outputs.
Whereas machines can be trained to detect and extract products of core word-building pro-
cesses, thereby facilitating the construction of large corpora, the forms of more informal
processes often must be compiled manually. (See, for instance, Gries’s (2004a) English
corpus of 585 forms touted as “one of the largest blend corpora analyzed so far.”) The
relative unpredictability of these forms is of direct interest but at the same time compli-
cates large-scale investigations and may thus lead to taxonomies and/or theories based on
comparatively few data.
Such processes, moreover, are not of purely empirical interest. In particular, they may
both evidence speakers’ implicit knowledge of phonological constraints or structures, in-
cluding those inactive in their native language, and allow us to argue for certain theoretical
constructs over others. As an example of the former, Tessier and Becker (2018) find that
a nascent class of insult compounds in English, shitgibbons, are judged as better formed
when the initial vowels, but not consonants, of the two source words are identical. This
tendency, especially in its asymmetry, can be established independently in other natural
language phenomena. As an example of the latter, Elfner and Kimper (2008) argue from
results on diddly-infixation (e.g., wélcome > wel-diddly-élcome) that phonological redupli-
cation need not be limited to the single segment or subject to rigid locality restrictions.
In this paper, I present another such class of words from a recent language game,
pussy blends (e.g., (Margaret) Thatcher + pussy > thatchussy), and examine their poten-
tial contribution to the study of informal word-building processes. As they appear at first
glance to be most similar to blends, the behaviour of 1,338 pussy blends in an automati-
cally gathered Twitter corpus are compared against the documented tendencies of blending
in the literature. Of particular interest is the amount and shape of material contributed by
each source word.
The present article serves as a preliminary report. That is, before we can determine
what the theoretical contribution of these forms may be, we need to establish what they
⇤
I would like to thank attendees of the annual conference of the Canadian Linguisti Association & the
Montreal-Ottawa-Toronto Phonology/Phonetics Workshop for their insight and commentary. Many thanks
to Zya staff, in particular Dylan, for insights into the Ditty app and the lifespan of the meme. Many thanks
also to Darius Kazemi, author of the @Ussybot, for reflections on the formation of these words. Any mis-
takes, as well as the interpretations of the meme and its elements presented here, are my own.

Actes du congrès annuel de l’Association canadienne de linguistique 2018.

Proceedings of the 2018 annual conference of the Canadian Linguistic Association.
c 2018 Michael Dow
 2

are and how they are formed. Ultimately, this study finds that so-called “pussy blends”
fit the profile of documented blends fairly poorly. (This term is, nevertheless, retained for
the remainder of the paper awaiting further research.) However, they do corroborate what
English speakers appear to know implicitly about syllabic structure in English.
The remainder of the paper is structured as follows: §2 summarizes the taxonomy
of blends and related forms, as well as the findings of several previous corpus studies on
blends. §3 presents the history of “pussy blends” within its popularizing context, that is,
the “one thicc bih” meme. The methodology involved in the preparation and treatment of
this study’s corpus is presented in §4. §5 presents the results, which primarily concern the
average grapheme and syllable contribution of these forms’ source words. A discussion
follows in §6, along with a conclusion and directions for future research.

2. Blending

Before discussing the history and characteristics of so-called “pussy blends,” it is worth-
while to consider the known properties of blends in the literature, especially in English.
The types of blends attested within and across the languages of the world are extremely
diverse, and as such, the questions and taxonomy arising from their study are complex.
Seeing as the main forms of interest in this paper are of a particular type, we shall mainly
focus our attention in this section on the behaviour of similar blends.
In its most basic form, blending involves the combination of two base lexemes. One
or both of the source words are frequently clipped, though not necessarily, depending on
the degree of overlap, e.g., German Paradies + Diesel > Paradiesel1 . Additionally, though
circumscribed examples exist (e.g., hypocritement + critique > hypocritiquement), blends
typically involve a linear juxtaposition. According to Fradin (2015) and Kemmer (2003),
this process often chooses as its switch-over point some shared material between the source
words, whether graphemic or phonemic or both. This tendency, however, may be language-
specific, as blends with overlap in Bertinetto’s (2001) English and Italian data amount to
only 45% and 24% of cases, respectively, versus those of German and French at 89% and
79%.
Other word-building processes share some of these properties, especially the juxta-
position of lexemes and truncation. Fradin (2015) further distinguishes blending via two
other properties: First, within a given language, it is unpredictable which of the two words
serves as the semantic head of the resultant blend. This criterion separates blends from
compounds, whose semantic interpretation is much more regular within a given language.
In addition, compounding rarely, if ever, involves truncation (e.g., Kemmer 2003).
Second, Fradin claims that blending produces “type hapaxes” in a language; that is,
any constituent part of a blend cannot freely combine with other lexemes. This distin-
guishes blending from secreted affixation, as in chocoholic or emailgate. Tournier (1985)
names elements such as -(a)holic and -gate (from Watergate) fractomorphemes. Beyond
1 Examples in this section, if not otherwise stated, come from Fradin (2015).
 3

the increased productivity of secreted affixation, it also involves specific semantic opera-
tions not involved in blending (Fradin, 2015). Similarly, though more literal in its inter-
pretation, concealed compounding (e.g., freeware, shareware) is excluded by Fradin on the
basis of its series-like nature.
Scholars distinguish within blends proper two different types: telescopes and port-
manteaux (Piñeros, 2004), or overlap and substitute blends, in Kemmer’s terminology. The
former function essentially as linear expressions (e.g. Spanish cuernos + nacionales >
cuernacionales; ‘horns’ + ‘national’ > ‘national horns’) whose non-converging edges al-
ways remain intact, and whose resultant form is always longer than either of the source
words in terms of segments (Piñeros 2004, following Algeo 1977). Meanwhile, the con-
tributing words of portmanteaux are selected more for a combination of semantic asso-
ciation and phonological similarity, e.g., ladrón + McDonald’s > ladrónald’s; ‘thief’ +
‘McDonald’s’ > ‘McDonald’s as a rip-off’. As these words tend to mirror closely the
phonemic and/or syllabic structure and length of one of their source words, it is argued
these are formed by overlapping (Pharies, 1987), or concealment (Piñeros, 2004).
The length of blends in comparison with their source words is a recurrent question in
the literature, one which interests us especially here. In terms of syllables, the claims are
varied, potentially due to differences in corpora. Kubozono (1990) claims that blends are
often identical in syllable count to that of their second source word, while Bat-El (2006) and
Cannon (1986) claim it is the longer source word that determines this number. Segmentally
speaking, several authors find a strong tendency for the shorter of source words to con-
tribute more material to blends (e.g., Bergström, 1906; Kaunisto, 2000). Gries (2004a,b)
confirms this observation, in addition to arguing for a general, if at times conflicting ten-
dency for the second word to contribute more.
The question of which word contributes more is not independent of other factors; in
fact, it has been shown to be intertwined with word frequency and prototypicality (Kelly,
1998), order (Kubozono, 1990) and phonological similarity (e.g., Bat-El, 1996; Kemmer,
2003; Gries, 2004a). Given that the order and similarity of the second word are controlled
for in the present corpus, however, these issues are set aside here.
The final aspect of blends addressed here concerns syllable shape of blends’ con-
stituent parts. Bertinetto (2001) finds a language-specific effect in lexical blends without
overlap (the type most representative of pussy blends). Namely, while English shows a
preference for “right-branching” types, in which the blend’s juncture is situated between
onset and rhyme, German, French and Italian prefer “neutral” types (CV-CV juncture).
Syllabic structure can also influence order, in that source words with complex onsets may
be preferred as the first word and those with complex codas as the second (Bat-El, 2006).
This concern is, again, secondary, as order is fixed in our corpus.

3. Background

The following section provides background on pussy blends within and outside of the “one
thicc bih” meme that popularized them. First, the origins of the main elements of the
 4

meme’s formula are discussed in §3.1. Section 3.2 then examines the content and trajectory
of the meme. Section 3.3 concludes with details on pussy blends after the zenith of the
meme and offers some speculation on the future of these blends.

3.1 Ancient history (in Internet Years)

All the main elements of the “one thicc bih” formula, i.e., “thicc,” “bih,” and pussy blends
predate the meme itself. The first two can be traced back to AAVE slang, being alternate
spellings for “thick” and “bitch,” respectively. The former refers to an ample or muscular
body which the speaker typically finds attractive. Though the general use of “thick” in this
sense is hard to trace, according to KnowYourMeme2 , the earliest known instance on the
Internet of the particular spelling “thicc” comes from a 2004 hip-hop fan page. Meanwhile,
entries for “bih” date back to 2003 in the Urban Dictionary3 .
Pussy blends are first noted in gay slang, particularly (if not limited to) the formations
boy + pussy > bussy and man + pussy > mussy. These are, again, difficult to pinpoint in
time. The website Robert Scott’s Gay Slang Dictionary contains both “boy-pussy” and
“man-pussy” but not their blended forms. Both were submitted by a London-based user
in July 1999.4 The earliest entries in Urban Dictionary date back to 2004 (though in the
earliest entry, defined as “butthole pussy”).5
The recent popularity in pussy blends can confidently be attributed to an April 2017
Tumblr post which coins thrussy from throat pussy, garnering over 26,500 “notes” (simi-
lar to likes) within 72 hours of publication. Neither Urban Dictionary nor Google Trends
shows hits for “thrussy” before April 2017. Before this, it is unclear to what degree pussy
blends existing between 2004 and 2017 reflect a freely combinatorial process of word for-
mation (i.e., if novel forms would have been accepted), or whether they belonged to a
closed class. For instance, the word “babussy” appears in 2012 in a Ukranian political
movement as a combination of “babushka + pussy” (in reference to the band Pussy Riot).6
Further research may elucidate these questions, but for the moment it is assumed that pussy
blending remained marginal and restricted to a few items until the advent of thrussy and
the “one thicc bih” meme.

3.2 You can’t get rid of the Babadook

The “one thicc bih” meme as we know it first appears towards the end of May 2017. The
earliest, most popular instances of the meme involve the characters Mr. Krabs from the
animated T.V. show Spongebob Squarepants and the Babadook, the eponymous monster
2 https://knowyourmeme.com/memes/thicc
3 https://www.urbandictionary.com/define.php?term=Bih

4 Accessible
via the Internet Wayback Machine: https://web.archive.org/web/20010410014457/
http://hurricane.net:80/~wizard/dict_frame.html
5 https://www.urbandictionary.com/define.php?term=bussy
6 https://blogs.wsj.com/emergingeurope/2012/08/17/they-are-calling-it-babussy-riot/
 5

of a 2014 Australian horror movie (blended as krussy and babussy, respectively). At the
time of writing in June 2018, these original posts’ videos have garnered over 576,000 and
829,000 views alone (including retweets and shares).7 On the heels of these notable exam-
ples, the meme quickly become popular enough to warrant articles in sources such as New
York Magazine, Buzzfeed, and so on.
The essential formula of the meme is as follows: “x is one thicc bih, let me see that
y,” where x designates a noun (typically a character or famous personality), and y rep-
resents a blend of that word, or a semantically related word, and the word “pussy”. In
some instances of the meme, the pussy blend directly references an orifice and/or sexu-
alized part of the referent x (akin to “let me see dat ass”), while others, especially novel
and rhyming/near-identical forms (e.g., pharmacy > pharmussy), employ meta-humour to
defy expectations and/or provide commentary on the subject or on the meme itself. The
semantics of pussy blends are complex and may warrant further research; for the moment,
we are most interested in their form.
The “one thicc bih” meme circulated notably, though not exclusively, on Twitter,
both in text and accompanied by a video generated by Ditty (Zya Music), a text-to-song
app. Text provided by users is dynamically fit to the melody of any of a number of songs
and sung by a computer-generated voice. The meme in question uses one of the app’s
basic songs, “Good Day” by 4Qent100. To illustrate how the algorithm adapts to inputs of
various lengths, table 1 breaks down three different examples (containing pussy blends of
two, three and four syllables respectively) into simplified groups of 4 beats.

Mi ster Krabs is one thicc bih

The Baba dook is one thicc bih
Ka ren is one thicc bih lemme
1 2 3 4 1 2 3 4

lemme see that kru ssy

lemme see that babu ssy
see that compu tu ssy
1 2 3 4 1 2 3 4 1
Table 1: Metrical adaptation of 3 memes of varying lengths by Ditty app

3.3 Life after thiccness

As can be extrapolated from the popularity of the phrase “one thicc bih” in Google trends
(shown in Fig. 1), worldwide interest in the meme begins to decline in late June and reaches
7 Because of the large number of pussy forms cited in this paper, in the interest of space, the URLs and authors
of individual posts are not cited but may be provided upon request. It should be noted, though, that certain
posts may have been deleted, or entire accounts may have been deleted or modified in the period following
data collection.
 6

its nadir in August 2017, after which time interest remains low but stable.

Figure 1: Google trends for “one thicc bih” (03/2017 - 02/2018)

While the meme’s lifespan was not unexpectedly short, it remains to be seen whether pussy
blends as an open class of words will follow suit—that is, whether speakers will continue to
invent new pussy forms, whether certain forms will persist, but as a closed class, or whether
they will all fade into obscurity.
Limitations of data-gathering currently make this question difficult to investigate, as
substring searches—that is, all words ending with “ussy”—are not allowed by the Twitter
API (or for that matter, most, if not all, API). Data may be gathered en masse and post-
treated, though initial attempts are showing such forms to be rare. Meanwhile, searches
for individual words prove more promising (e.g., a spike in grussy following the release of
the trailer for the 2018 The Grinch film), though the process proves more time-consuming
and less exhaustive. The durability of pussy blends may ultimately need to be investigated,
along with their interpretability and combinatorial freedom. For the moment, it is assumed
that the phenomenon is still fresh in the public memory and therefore worth investigating.

4. Methodology

In order to examine the factors potentially driving pussy blend formation, a corpus was es-
tablished along the following lines. The Python script GetOldTweets8 was used to scrape
all publicly available tweets containing the phrases “one thicc bih” and “see that” published
between June and August of 2017. (The reader is reminded from §3.3 that the reason for
using this meme as a framing context is the impossibility of substring searches on this
platform.) Though the sources presented earlier in §3.2 suggest that the initial rise in popu-
larity of the meme occurred in May 2017, the script could not gather tweets from this time
because of limitations of the free Twitter API. The earliest date represented in the entire
corpus is June 17, 2017; however, as only a subset of the data was processed for the present
study, and this done by date in descending order, the earliest date represented in this paper
is June 23, 2017 (the latest date being August 30).
8 https://github.com/Jefferson-Henrique/GetOldTweets-python
 7

After exclusion of retweets and identical tweets, the corpus contains 4,450 tweets,
while the subset presented here is based on the 1,500 most recent tweets. The following
other information, beyond the text itself, was automatically gathered: date of publication,
username, number of retweets and favourites, unique identifier and URL.
The subset of 1,500 tweets (hereafter referred to as the corpus for the sake of brevity)
was further processed to extract pussy blends and their intended referents. First, the words
immediately following “see that” up to the substring “ssy” (or the end of the line in absence
of this ending) was automatically extracted and manually inspected. Forms which could
not be conclusively identified as a pussy blend were targeted for elimination. These fall
into three types, and though coincidentally the first two are not found in the subset corpus,
they are all illustrated here with examples from the larger corpus. First of all, formations
unambiguously using other words than “pussy” as the second element of the blend (e.g.,
(Don) Draper > Drapenis) do not qualify. Second, formations such as Vlad > Vladdy and
wolf > wolfy cannot be reliably distinguished from diminutives and thus do not qualify
either. Finally, tweets using existing words which cannot be confidently separated into
constituent source words were also eliminated. A popular version of this type is illustrated
by the tweet “Claire de Lune is one thicc bih, let me see that Debussy”, in which the
name “Debussy” cannot be separated into “pussy” and a “Deb(u)–”-initial word. Such
cases accounted for 38 of all excluded forms. Note that forms appearing to use alternate
spellings of “pussy” (e.g., Mermaid > mermsey) in the blend were not present in the subset
but would not have been excluded. Finally, the pussy blend was standardized to remove
effects of orthographic lengthening (e.g., “HarLussYYyyyY” coded as “harlussy”).
The referent of the meme was then automatically extracted as the words preceding
“is one thicc bih” (variants of the copula, such as “iz,” included) and again were manually
verified. These referents were then compared with the pussy form to establish both the full
first source word and the novelty of the pussy blend. In non-novel cases, the source word
was often unambiguous, though certain conventions had to be adopted. For instance, full
names were taken as the source word only when all words were included without a space
in the pussy blend (e.g., Ms. Puff > mspussy coded for “mspuff”); otherwise, only the
word explicitly contributing information was tagged (e.g., Street lego man > legussy coded
for “lego”). Note that any punctuation and spaces in the full referent were removed in the
coded source word. In rare cases of potential ambiguity where two or more words of a
full name or expression began with the same grapheme(s), the last word was categorically
chosen (e.g., “cheese” taken as the base in Chuck E. Cheese > chussy).
The novelty of pussy forms was established at this same stage. Forms not taking the
explicitly stated reference as its first source word were defined as novel with an important
exception. Namely, unexpressed source words which can be interpreted as inherent to the
referent, which was limited to first/last name or species—for instance, Gary (a snail on
Spongebob Squarepants) > snussy—were counted as non-novel. These were, however,
given a special code and can thus be reclassified later if necessary. The source words of
novel forms were finally reconstituted based on the conversation thread, the user’s Twitter
page, and Google searches. Ultimately, 124 forms had to be discarded for ambiguity. All
 8

in all, 162 forms were discarded for a total of 1,338 pussy blends.
Source word contributions to the blend were then manually coded. Following Gries
(2004b), the source word contributions were formatted for a generous interpretation, though
strictly locally within each source word. That is, the number of graphemes contributed by
syllabus and pussy in syllabussy would be 8 and 4 respectively (i.e., <syllabus> + p<ussy>).
A hypothetical example of Jerusalem > jussy would not be coded as <j>er<us>alem +
p<ussy>, while jerussy would be coded as <jerus>alem + p<ussy>. This convention is
likely to be complicated when phonemes, rather than graphemes, are considered. The
question is left open for the time being.
No single user is overwhelmingly represented in the corpus.9 The current 1,338
tweets were composed by 1,156 unique users. The max number of forms provided by
a single user is 6, while the mean is 1.2 tweets per user. Variation across users is thus
assumed in the results rather than directly incorporated.
In order to calculate syllables in source words and blends, vowels were counted in
the following way. Contextual digraphs were first processed, e.g., the <i> in <tion> was
excluded, and then vocalic digraphs were transformed into single nuclei. Then all single
vocalic graphemes were tabulated. The resulting syllable counts were manually verified for
all 1,338 forms.
Finally, the Similarity Index of each blend was calculated according to Gries’s (2004b)
formula. That is, the relative number of graphemes maintained by each word was multi-
plied by the relative number of graphemes of the blend for that word. These scores were
added and divided by two. The formula is provided in table 2, where G = no. graphemes, r
= root (i.e., source word), c = contributed material, and b = blend.

(G c1 Gc1 Gc2 Gc2

Gr1 ⇥ Gb ) + ( Gr2 ⇥ Gb )
2

Table 2: Similarity Index equation (Gries, 2004b)

For example, <syllabus> + p<ussy> yields an SI of 0.56, corresponding to well-formed

blends in Gries’ study.

5. Results

This section presents some preliminary findings from the study thus far, preceded by some
anecdotal remarks from the data. The SI results are then reported along with word con-
tributions by syllable size and shape. Given that, as we will see, the amount of material
contributed by the second word is near-invariable, we are interested for the moment in the
factors determining the amount of material contributed by W1 .
9 The robot account @Ussybot, which automatically generates pussy blends, was intentionally excluded from
the data. It cannot be guaranteed that the corpus contains no robots, but if present, they are assumed to be
extremely few.
 9

5.1 Preliminary remarks

The forms discussed in this section are more often than not rare, in comparison with those
presented in §5.2, though it remains to be seen whether or not they are truly exceptional.
First, several examples of (near-)inclusive blending (Ronneberger-Sibold, 2006) can be
noted in the data. This type of blend results in a form whose pronunciation is the same
as the longer source word but whose other source word can be recovered from the blend’s
written form, e.g., Jew + jubilee > jewbilee, an example from the corpus being democracy
> democrussy. Essentially, these amount to cases with a [-@sI]-final W1 .
Only a small number of forms take more than two source words, e.g., octoling choco-
late + pussy > octochussy. It is unclear how these should be treated, namely whether the
derivation should be flat (e.g., <octo>ling <ch>ocolate p<ussy>) or recursive (<ch>ocolate
p<ussy> ! <octo>ling <chussy> or <octo>ling <ch>ocolate ! <octoch> p<ussy>). For
the moment, they are treated linearly, the same as 2-word combinations, as combinations
of 3+ words are very few in number.
A handful of forms evidenced intrusive consonants, that is, material belonging to
neither source word. These largely included <r> (e.g., Fionn > Fiorussy) and <b> (e.g.,
Jake > Jabussy) and can arguably be based in analogy to thrussy and bussy. A similar but
more inexplicable type involves initial-consonant copy (e.g., Gao > gagussy, Cameron >
Cacussy, me > memussy).

5.2 Syllables and graphemes

Here, we present a number of calculations in order to paint a portrait of the average blend in
the corpus. Forms are generally non-novel (n = 1,187, 88.7%), and only 154 forms (11.5%)
have graphemic overlap (only 15 of which are novel). Hiatus is largely avoided, only 114
blends (8.5%) containing a V-V juncture. As for the W2 contribution, <ussy> is by far the
most common (1,238, 92.5%), with <pussy> and <ssy> tied at 50 forms (3.7%) each. The
mean SI of non-novel forms is 0.37, while that of novel forms reaches 0.42.
As shown in table 3, there is an overwhelming preference in non-novel forms for
a monosyllabic W1 contribution (that is, considering the syllabification of only the con-
tributed material). Novel forms share equal preference for mono- and disyllabic contribu-
tions. Non-novel forms prefer onset-only contributions to disyllabic shapes.

Condition Onset-only 1s 2s 3s 4s 5s
Non-novel 286 719 151 27 1 0
Novel 8 61 63 13 3 2

Table 3: No. forms per W1 syllable contribution by novelty

Expanding on this, we can see in table 4 that there is a general preference for W1 con-
tribution to end in a single consonant, regardless of the number of syllables contributed,
especially in the non-novel forms. Note that in table 4, only the number of consonants at
 10

Count Count
Shape
(Non-novel) (Novel)
C1 208 3
C2 108 6
CV 62 3
CVC1 466 26
CVC2 201 32
CVCV 31 6
CVCVC1 63 42
CVCVC2 21 15

Table 4: W1 contribution syllable shape by novelty

Condition Gr (W1 ) s (W1 ) Gr loss Count

W1 < W2 2.5 0.69 1.2 235
Non-novel W1 = W2 2.9 0.66 2.1 318
W1 > W2 3.8 1 3.5 634
W1 < W2 3.4 0.9 0.5 23
Novel W1 = W2 4.1 1.2 0.9 33
W1 > W2 6 1.9 1.9 95

Table 5: W1 contributions and loss by novelty & relative length

the end are considered. For instance, CVC1 could refer to VC, CVC or CCVC contribu-
tions, but CVCC is necessarily distinguished as CVC2 . Given that 3+ consonant sequences
are rare, C2 refers to 2 or more consonants at the end of W1 contribution. Finally, CV does
not distinguish <u>-final W1 from cases which would lead to true hiatus. Contributions
greater than 2 syllables are excluded. As for the sonority profile of C2 -final syllables, there
is a marked preference for falling sonority (161 of 272, 59.2%), then rising (84, 30.9%)
and even (27, 10%).
So far the results have considered grapheme and syllable contribution regardless of
the original length of the source words. Table 5 shows the mean contribution of W1 (that
of W2 being constant at 4 graphemes) according to the relative lengths of the two source
words, separated by novelty. Not unexpectedly, within the same condition, W1 contribution
is greater in novel blends. Within non-novel and novel blends, W1 contribution increases
across conditions; that is, the first word contributes the most material when longer than the
second. These are the majority of cases, regardless of novelty, though note that in the case
of non-novel blends, the result is effectively equal contribution as W2 (3.8 vs. 4).
 11

Word 1 contribution 3

1 2 3 4 5 6
Source word 1
Non−novel Novel

Figure 2: No. syllables, source word vs. contribution (W1 )

Finally, we consider the interaction between the number of syllables provided by W1

against the source word’s original length, as shown in figure 2. In the non-novel condition,
contribution clusters around 1 syllable, while contribution in novel blends starts on average
higher and appears to increase proportionately with source word length.

6. Discussion and Conclusion

The questions asked in this report are twofold: First, what factors, presumably phonolog-
ical, influence the amount and shape of material contributed in so-called “pussy blends”?
Second, can they rightfully be considered blends?
To remind the reader, we know the factors determining the resultant blend of any two
words to be multifaceted, among which are phonological similarity, word order, relative
length, language-specific parameters, and so on. The very nature of the present language
game allows us to isolate just a few. Namely, word order is fixed, as is (largely) W2 con-
tribution, and phonological similarity is not a prerequisite. We can thus focus on what
attributes of a given word, in relation with the word “pussy,” may determine the material
contributed to their combination.
In non-novel conditions, W2 contributed more on average, regardless of relative length.
This is slightly in keeping with one of the observations in Gries (2004b), where it is
claimed:

“. . . x segments of the beginning of a word increase its chance of being recog-

 12

nised more than x segments of its end (cf. Nooteboom 1981), given that the
normal way we encounter words is from beginning to end rather than vice
versa. Therefore, it makes sense that, if both source words are equally long,
the second word contributes more because this would enhance its recognisabil-
ity by compensating for the fact that it is not processed in the normal way.”
In another study, though, using tests of statistical significance, Gries (2004a) finds source
words contribute equal material when equally long. This he attributes to phonological
similarity and overlap counteracting the above mentioned findings. Similarity and overlap
are largely absent from this corpus and thus cannot be invoked.
Gries’s other observation, a competing constraint that the shorter word tends to con-
tribute more material (see also Kaunisto 2000), takes precedence outside of this condition;
indeed, the strongest factors in Gries (2004a) are that the shorter words contribute more,
regardless of position. Again, this is not borne out in the present data in the non-novel
condition.
Meanwhile, in the novel condition, W2 contributes more when it is longer, contributes
equally when the two are equal, and W1 contributes more when it is longer. The equal
length data match Gries (2004a), but otherwise the results skew in the opposite direction;
that is, the longer word contributes more.
Relative length matters not only in the literature on blends as a whole, but also in
the subcategorization of blends, i.e., telescopes vs. portmanteaux. In some respects, pussy
blends are clearly closer to telescopes than portmanteaux. The relative lack of overlap, es-
pecially in novel forms, suggests that the interaction of phonological overlap and semantic
association (effectively amounting to the wit or felicity more characteristic of portman-
teaux blends) plays a minor role, if at all, in pussy blends. Rather, their parts stand in a
syntagmatic relation, “x + pussy” essentially meaning a sexualizable part of x. However,
the non-negligible presence of C+ussy forms (15.7% of the corpus) goes against Piñeros’s
(2004) claim that telescopes are always longer than their source words. In fact, more than
half the C+ussy forms (112 of 211) are equal to or shorter than W1 in addition to W2 .
Moreover, beyond these shorter forms, as seen in tables 3, 4 and 5, there may be
evidence in the non-novel condition for a CVC(C) templatic contribution from W1 . We see
this both in the number of closed monosyllabic contributions and more subtly suggested
when we consider the relative lengths of each word. (The effects of stress, though, must
be sorted out, especially in these longer words.) This runs counter to Bat-El’s statement
that “the number of syllables in a blend is often identical to the number of syllables in
the longer base word” (Bat-El, 2006, 67). Pussy blends may, in fact, be maximally, if not
ideally trisyllabic. (Ideally in the sense that, especially context of the “one thicc bih” meme,
few disyllabic blends are likely to be distinct enough.)
Speaking of syllables, let us briefly turn our attention to the relative shapes of “pussy
blend” source word contributions, in lieu of Bertinetto (2001). As argued earlier, W1 con-
tributions more often than not end in at least one consonant, regardless of the number of
syllables preceding. Contribution-final CC sequences also tended to fall in sonority. Mean-
while, W2 is almost categorically a vowel plus an extra CV syllable (i.e., -[Usi]). These
 13

correspond in Bertinetto’s (2001) system to Onset + Rhyme and Closed syllable + Rhyme
types (the latter in the case of falling CC sonority sequences). These right-branching types
were the most frequent type in English (though not cross-linguistically).
To summarize, what have been called “pussy blends” largely fail to convene to most
aspects of blends. In terms of relative length and contribution, they rarely match the trends
observed in previous corpus studies, at times reversing them. Support that these forms can
be categorized as telescopes is questionable; their formation does not seem to follow that
of true telescopes. Finally, their series-like nature constitutes a technical but immediate
disqualification for Fradin (2015).
In sum, “pussy blends” are more likely to fall into the category of secreted affixa-
tion, in which case -ussy would have become, however ephemerally, a fractomorpheme (cf.
Watergate > -gate). Further supporting this is the seemingly regular semantic operation
transforming the meaning of “pussy,” in the same way that the meaning of “Watergate” is
transformed into “scandal connected with x involving high-ranking people” (Fradin, 2000).
To my knowledge, no large-scale investigation focussing on secreted affixation has
been conducted. As such, the forms investigated here (and to be investigated in the larger
corpus) may reveal more nuanced structural differences between these and blends, beyond
the well-established semantic differences. In the future, judgement tasks and surveys from
meme creators and/or consumers may also allow us to further elucidate open questions, in
particular the following: whether stress can influence the switch-over point, whether W1
contribution is indeed ideally a monosyllabic template and whether this word-formation
process, or even certain of its forms, is expected to remain in the public eye or disappear
into the annals of Internet history.

References

Algeo, John. 1977. Blends, a structural and systemic view. American Speech 52: 47–64.
Bat-El, Outi. 1996. Selecting the best of the worst: The grammar of Hebrew blends. Phonology 13:
283–328.
Bat-El, Outi. 2006. Blend. In Encyclopedia of Language & Linguistics, ed. Keith Brown. Elsevier,
66–70.
Bergström, Gustav A. 1906. On blendings of synonymous or cognate expressions in English: a
contribution to the study of contamination. H. Ohlsson.
Bertinetto, Pier Marco. 2001. Blends and syllabic structure: A four-fold comparison. La gramàtica
i la semàntica en l’estudi de la variació : 59–112.
Cannon, Garland. 1986. Blends in English word formation. Linguistics 24(4): 725–754.
Elfner, Emily and Wendell Kimper. 2008. Reduplication without RED: Evidence from diddly-
infixation. In Proceedings of the 27th West Coast Conference on Formal Linguistics. Cas-
cadilla Press Somerville, MA, 150–158.
Fradin, Bernard. 2000. Combining forms, blends and related phenomena. Extragrammatical and
Marginal Morphology 12: 11.
Fradin, Bernard. 2015. Blending. In Word-formation. An International Handbook of the Languages
 14

of Europe, ed. Peter O. Müller, Ingeborg Ohnheiser, Susan Olsen, and Franz Rainer. Berlin:
De Gruyter Mouton, 386–413.
Gries, Stefan Th. 2004a. Shouldn’t it be breakfunch? A quantitative analysis of blend structure in
English. Linguistics : 639–668.
Gries, Stefan Th. 2004b. Some characteristics of English morphological blends. In Papers from the
38th Regional Meeting of the Chicago Linguistics Society, Vol. 2. 201–216.
Kaunisto, Mark. 2000. Relations and proportions in English blend words. Fourth Conference of the
International Quantitative Linguistics Association.
Kelly, Michael H. 1998. To “brunch” or to “brench”: Some aspects of blend structure. Linguistics
36(3): 579–590.
Kemmer, Suzanne. 2003. Schemas and lexical blends. In Motivation in Language: Studies in
Honor of Günter Radden, ed. Hubert Cuyckens, Thomas Berg, René Dirven, and Klaus-Uwe
Panther, vol. 243. John Benjamins Publishing, 69–97.
Kubozono, Haruo. 1990. Phonological constraints on blending in English as a case for phonology-
morphology interface. Yearbook of Morphology 3: 1–20.
Nooteboom, Sieb G. 1981. Lexical retrieval from fragments of spoken words: Beginnings vs.
endings. Journal of Phonetics 9(4): 407–424.
Pharies, David A. 1987. Blending in Spanish word-formation. Romanistisches Jahrbuch 38: 271–
289.
Piñeros, Carlos-Eduardo. 2004. The creation of portmanteaus in the extragrammatical morphology
of Spanish. Probus 16(2): 203–240.
Ronneberger-Sibold, Elke. 2006. Lexical blends: Functionally tuning the transparency of complex
words. Folia Linguistica 40(1-2): 155–181.
Tessier, Anne-Michelle and Michael Becker. 2018. Vowel but not consonant identity and the very
informal English lexicon. In Proceedings of the 2017 Annual Meeting on Phonology. 1–12.
Tournier, Jean. 1985. Introduction descriptive à la lexicogénétique de l’anglais contemporain.
Champion Books.