How novel are low-frequency words? Investigating the usage of infrequent non-lexicalised Xable adjectives in contemporary English using the NOW corpus Chris A. Smith This study explores the relationship between low-frequency lexical forms and lexical innovation by examining infrequent and non-lexicalised adjectives formed with the suffix -able (e.g., jokeable, trickable). Employing a corpusbased analysis of the 20-billion-word News on the Web (NOW) corpus (2010- 2025), we identified over 800 infrequent Xable adjectives. After filtering for orthographical errors, brand names, and incorrect forms, a dataset of novel lexical items was isolated using the OED for evidence of attestation. Morphological patterns and usage contexts were analysed, highlighting factors such as (1) frequency distribution over 2010-2025, (2) geographical coverage, (3) contextual anchoring in the collocate environment, and 4) base form and derivative co-occurrence. Findings contribute insights into lexical innovation, morphological productivity, and the stages of lexical integration from a dynamic usage-based perspective (Schmid 2020). 1 1. Introduction: concepts and framework 1.1 Neologisms Considerable scholarly work has been devoted to lexical innovations and neologisms in the field of lexicology and word formation (Tournier 1985; Baayen 1991; Baayen & Lieber 1997; Baayen & Renouf 1996; Bauer 1983; Schmid 2008, Schmid 2011, 2014, 2015; Miller 2014; Kerremans 2015; Mattiello 2017; Anesa 2019). These studies typically explore lexical nov- 1 I thank the anonymous reviewers for their helpful feedback and guidance. AAA - Arbeiten aus Anglistik und Amerikanistik Agenda: Advancing Anglophone Studies Band 51 · Heft 1 Gunter Narr Verlag Tübingen DOI 10.24053/ AAA-2026-0002 Chris A. Smith 8 elty, morphological innovation, and semantic shifts utilising usage-based approaches. Advancements in cognitive linguistics and psycholinguistics, particularly in Entrenchment and Conventionalisation Theory (Schmid 2020), Construction Grammar (Goldberg 2019; Hoffman 2020), and Construction Morphology (Booij 2010, 2018; Booij, Jackendoff & Audring 2016), have significantly reshaped the field in the past 15 years. Neologisms are inherently transitional and diachronic, representing different stages of lexical integration (Schmid 2008; Mattiello 2017). Kerremans (2015) categorises neologisms into three types: morphological, formed through new word structures (e.g., snowicane); semantic, where existing words gain new meanings (e.g., subprime); and morphosemantic, combining both processes (e.g., detweet). The study of lowfrequency lexical items is essential for understanding morphological productivity and the broader processes of lexical change. Productivity, broadly conceived, designates the potential of word-formation processes to coin new words and thus contribute to ongoing lexical innovation (Aronoff 1976; Baayen & Lieber 1991; Plag 1999; Bauer 2001; Baayen 1992; Huygue & Varvara 2023). These words provide a unique perspective on the evolving nature of word formation, revealing patterns of innovation, stabilisation, and eventual obsolescence. Baayen & Renouf (1996: 83) highlight the significance of low-frequency words in assessing the productivity of morphological rules, suggesting that their presence may indicate either the overextension of a productive process or its gradual decline. In the latter case, such words serve as linguistic fossils - remnants of onceproductive formations that are now receding from common usage. An illustrative example is winsome (‘charming’), which originates from Old English wynn (‘joy’) combined with the adjectival suffix -some. While some was once a highly productive suffix in forming adjectives (e.g., troublesome, fearsome), its productivity has greatly diminished in contemporary English. As a result, winsome persists primarily as a low-frequency, lexicalised remnant - functioning as a linguistic fossil of an earlier stage in the language’s morphological history (see Smith 2024). This decline often occurs in contexts of morphological competition (Bauer 2006), where two or more affixes compete to fill the same semantic or functional niche. As Arndt-Lappe (2014) and Huyghe and Varvara (2023) observe, such rivalry can limit the productivity of one form while favouring another, leaving behind residual formations that persist at low frequencies. Data-driven studies on diachronic competition, including low-frequency items, such as Smith (2020) on Xsome and Xable adjectives, highlight the need for an effective productivity model to explain the interaction between competing derivational patterns. How novel are low-frequency words? 9 1.2 Productivity of Xable Morphological productivity is typically measured through the prevalence of low-frequency items, particularly hapax legomena (words appearing only once), rather than mere output size. Numerous studies have established that morphological productivity is typically associated with lowfrequency forms, whereas high-frequency forms are more often the result of lexicalisation (Bauer 1983: 61; Bauer 2001: 47-49; Jackendoff 1997: 231; Plag 1999: 24-34; Aronoff & Anshen 2001; Lipka 2002: 111; Hohenhaus 2005). The -able suffix is a particularly productive and widely studied component of English morphology due to its ability to generate a vast range of adjectives from verbal (and nominal) bases (Aronoff 1976; Bauer 1983, 2005; Plag 2003; Lehrer 2004; Oltra Massuel 2014; Miller 2014). Typically, -able attaches to transitive verbs to form adjectives denoting capability or potentiality (manageable, readable). However, the pattern exhibits considerable variation, including attachment to particle verbs (laugh-at-able) and compound or complex structures (car-toppable). Additionally, -able can affix to nouns, as seen in marriageable (attested in 1585 in the Oxford English Dictionary, henceforth OED). From a construction morphology perspective, the X-able pattern consists of an open slot X, which can be occupied by verbs, nouns, and potentially other lexical categories. In schema-based models of word formation, schemas function as structured templates containing variable slots (Booij 2010: 41-43). These schemas account for the productivity, analysability, and capacity for extension (or productivity) of morphological patterns (Booij 2010; Bybee 2010; Schmid 2015; Booij & Masini 2018; Jackendoff & Audring 2016: 471). Empirical approaches to morphological productivity have commonly utilised corpus frequency, particularly hapax legomena, as key indicators (Baayen & Lieber 1991; Baayen 2009). Register and textual context significantly influence morphological productivity, and dictionary-based measures often underestimate productive morphological processes like able suffixation (Baayen & Renouf 1996; Bauer 2001). To enhance the validity of purely quantitative measures of productivity, Goldberg (2019: 63) argues for incorporating an assessment of coverage, defined as “the degree to which the attested instances fill or ‘cover’ the category” of the designated slot. This approach ensures that productivity is evaluated not only by the number of occurrences but also by the extent to which a morphological pattern systematically applies across a given category. Chris A. Smith 10 1.3 Cognitive factors in lexical innovation Cognitive and psycholinguistic dimensions influence how new lexical items diffuse and integrate into language use. Contextual anchoring - both morphological and semantic - facilitates comprehension and storage of neologisms within mental lexicons (Kerremans 2015; Baayen & Neijt 1997). Lexical storage thus reflects a continuum between complete lexical exemplars and abstract morphological patterns or schemas, although there is some ongoing discussion of what is stored (Schmid 2015; also Jackendoff & Audring 2016). Consider the low-frequency adjective murderable, which occurs a mere 9 times in NOW, as in (1), where murderable refers to a psychopathic character from the cult TV series Game of Thrones: (1) And if you’re looking for silver linings, there are plenty -by Ramsay effectively seizing Winterfell, it now makes him the North’s most murderable. (US, 16-05-02) Despite its very low frequency, murderable (from murder V ) is attested in the OED as early as 1920 in the sense ‘giving cause for murder; liable to be murdered; provoking or inviting murder’. Low frequency forms can tell very different stories: they may or may not be attested in dictionaries and their usage can correspond to different stages in the life cycle of a form (Anesa 2019: 41). As shown in Figure 1, lexical life cycles are far from linear (which is supported by Smith (2024) in a study of lowfrequency -some derivatives). Figure 1. Potential stages of lexical cycle in Anesa (2019: 41) 1.4 Research objective and lifecycle model This corpus-based study investigates low-frequency -able derivations, drawing from the NOW (News on the Web, 2010-2025) corpus (Davies How novel are low-frequency words? 11 2016), and cross-referencing these with the OED to assess lexical innovation and establishment. Kerremans (2015: 40) examines the process of lexical establishment, emphasising that newly coined lexemes do not immediately form stable mental representations but are initially processed through their constituent parts. With repeated exposure, temporary entries are created in the mental lexicon and gradually consolidated through paradigmatic and syntagmatic associations (Schmid 2011: 76). Schmid (2008: 3) illustrates this establishment process from three complementary perspectives: lexicalisation, institutionalisation/ conventionalisation, and hypostatisation/ entrenchment. Here, hypostatisation refers to the perception that a word implies the existence of a corresponding concept. Taken together, these perspectives frame lexical establishment as a three-stage process. Kerremans (2015: 115) identifies three indicators of the nonconventionalisation stage which is of specific interest here: (1) limited diffusion across source types; (2) limited occurrence across discourse fields; (3) low frequency in linguistic data. Consequently, following Kerremans (2015: 115), lexemes that fail to occur frequently, appear in fewer than two source types, or remain restricted to a narrow range of discourse fields, are strong candidates for non-conventionalised status. The threshold of ‘two’ reflects the assumption that occurrence in more than one source type indicates at least minimal cross-context adoption. Additionally, topicality - defined by Kerremans (2015: 122), following Fischer (1998: 16), as the use of a word in connection with current affairs for a short period of time - plays a crucial role in lexical transition, since momentary frequency spikes often correspond to specific extralinguistic events. Such lexical items may exhibit sudden bursts in usage across multiple sources before either stabilising or fading from use. The adjective murderable, attested in the OED from 1920 onwards and placed in frequency band 1 (below 0.01 occurrences per million words), appears to have consistently remained in low-frequency use according to the OED’s diachronic frequency indications. 2 This study aims to investigate the extent to which neologisms can be identified through a data-driven analysis of low-frequency lexical forms generated by the productive morphological process X-able. 2 A search for murderable in COHA (1820-2010) reveals no occurrences. Chris A. Smith 12 2. Corpus methodology 2.1 Corpus and data collection 2.1.1 Data extraction from NOW We began with a corpus-based search for potential -able neologisms in the NOW corpus, focusing on low-frequency forms defined as occurring at or below 0.01 pmw (≈20-30 tokens). This threshold, aligned with the lowest OED frequency band, balances the inclusion of genuinely rare items while avoiding the noise of hapax legomena. Similar cut-offs are used in corpus studies of morphological productivity (Baayen 1992; Plag 1999), and are suitable for a corpus of NOW’s scale, where even low relative frequencies yield several dozen tokens over 15 years. This search produced over 800 candidates. The number of high-frequency lexical items is relatively small, whereas the diversity of low-frequency words increases significantly as frequency decreases, a pattern described by Zipf’s law. This pattern complicates the identification of genuinely novel forms, as simple string searches prioritise higher-frequency items, making the extraction of very lowfrequency words inefficient and time-consuming. This observation aligns with Baayen and Neijt (1997: 568), who observe that frequently used lexical items are often well-established and exhibit morphological irregularities, whereas highly productive morphological patterns generate numerous low-frequency lexical items. Among the identified forms, one notable hapax was discovered: tinsurable, initially recorded with 28 occurrences in the NOW corpus, but upon closer inspection, it represented a single unique context in (2): (2) Launching one of the brand’s most exciting digital campaigns of the year with long-running social agency partner, Bounce, Brookvale Union promises that good times and good tins across the country are now protected: Any moment with a tinnie in hand is, in fact, ‘tinsurable’. (AU, 23-11-06) The campaign slogan in (2) demonstrates common characteristics of hapax legomena, including metalinguistic commentary via the use of quotation marks and contextual anchoring via the presence of morphological priming and accentuation with italics (tins and tinnie). To analyse these features, we used a semi-automatic extraction procedure followed by manual verification. We first generated an initial dataset and sorted the items by frequency, then manually checked each candidate in the OED to confirm its first attestation. This step was feasi- How novel are low-frequency words? 13 ble given the dataset’s size: 197 -able adjectives after removing spelling errors and forms containing additional affixes. The results reflect searches conducted in September 2024, noting that the corpus grows continuously (7-10 million words added per day). Corpus Size Data source Text genres Tools NOW corpus (News on the Web) created by Mark Davies 19.7 billion (on 2024- 09-14) Data from web-based newspapers and magazines from 2010 to the present time growing by or about 3.1 billion words each year. Monitored newspaper corpus Collocate search (MI) Text type search Short diachrony over 15 years EnTenTen21 52 billion (over 2x size) Texts were downloaded in October- December 2021 and January 2022. Web crawler including all information on the internet Collocate search (word sketch) Snapshot view of 2021 Table 1. NOW corpus composition compared to EnTenTen21 Ultimately, the data collection method will be subjected to scrutiny to check if it is possible to optimise both accuracy (targeting novel -able words) and precision (collecting only -able words) in the corpus. 2.1.2 Sorting process and results A total of 163 cases showed orthographic variation, largely corresponding to “silent morphology” errors (Fayol, Largy & Lemaire 1994), including common misspellings (e.g., destructable for destructible, maleable for malleable, winable for winnable), alternation between -ible and -able, retention of a silent e (eable/ able), and hyphenation. Manual refinement then isolated three main morphological patterns: - The largest group consisted of 216 Xable forms with negative prefixes (un-, in-, non-, dis-). This pattern aligns with Baayen and Renouf’s (1996) observation of the high productivity of unwith adjectives, especially deverbal -able forms. Prefixes such as un-, nonand antiwere frequently attested, confirming the productivity of Chris A. Smith 14 negative morphological constructions. A full theoretical discussion is beyond the scope of this section, but it is worth noting that certain items (e.g. non, anti) occupy a borderline position between prefixes and independent morphemes; for discussion of such classification issues, see Bauer, Lieber and Plag (2013) and Booij (2010). - The second largest group comprises 164 prefix + Xable forms. These include intensifying elements (e.g., highly-, hyper-, ultra-, uber-, extra-, over-, all-), approximative ones (e.g., near-, almost-, less-, just-), and temporal elements (e.g., ever-, still-, once-, oft-, always-, then-). Some of these items occupy a borderline position between prefixes and free morphemes, but are grouped together here for analytical convenience. Adverbials such as easily and seemingly are also included. - The third group consists of 66 N + Xable compounds, largely organised around productive semantic patterns. These include N + capable formations (e.g., Alexa-capable, voice-capable, cloudcapable, wifi-capable) and AGENT-oriented schemas (e.g., lighttriggerable, machine-readable, hand-holdable, wrist-wearable), many of which are technical or specialised terms (e.g., heat-shrinkable, air-droppable). Following these exclusions, 197 genuinely low-frequency Xable adjectives were retained for detailed analysis. Morphological schema Types Negative forms 216 Orthographic errors 163 Prefixed forms 164 Compounds (N + Xable) 66 Other items (retained for analysis) 197 Total 806 Table 2. Composition of low frequency dataset Xable in NOW The dataset also includes several brand-name uses of -able; five such forms (Table 3) appear in the final set of unlexicalised adjectives. 3 3 A further -able brand name was also found in the lower frequencies below 20 tokens Getaboutable. How novel are low-frequency words? 15 Brandname Tokens SENSABLE 30 BOXABLE 23 CAMPABLE 30 FLEXABLE 27 RETIRABLE 26 Table 3. Brandnames in -able in NOW dataset Admittedly, brand names are often deliberate, consciously crafted coinages, differing from the more spontaneous formations of everyday language. Even so, they can mirror emerging morphological tendencies and thus provide valuable insights into ongoing patterns of word formation. 2.1.3 Method using OED All 197 forms were cross-referenced in the OED. Unlike Baayen and Renouf (1996), who later argued that dictionary checks are unnecessary for hapax legomena, our focus on low-frequency - not single-occurrence - items made verification essential. Although dictionaries tend to underrepresent neologisms (Baayen & Renouf 1996: 69), most lowfrequency -able adjectives identified in September 2024 were already attested in the OED: only 66 of the 197 lacked an entry and were classified as potential neologisms. The presence of over one hundred lowfrequency yet dictionary-attested forms partly reflects underestimation of corpus frequencies due to orthographic variation, which we normalised using wildcard searches. More broadly, the coexistence of many lowfrequency items with clear dictionary attestation illustrates a common feature of morphological productivity, where productive patterns generate numerous rare forms (Baayen & Neijt 1997: 568; Baayen 2009). The final set of 66 unattested forms was systematically checked against the OED, confirming their status as unlexicalised and potentially novel items. These forms were then ranked by frequency (as of February 2025) and constitute the dataset used for the detailed analysis of morphological innovation. Variants included FREQ TOTAL Base word Variants included FREQ TOTAL Base word monetiz*able 1458 monetise compromisable 31 compromise blendable 1276 blend donat*able 31 donate crav*able 1124 crave fieldable 31 field bik*able 677 bike archiv*able 30 archive glanc*able 528 glance campable 30 camp Chris A. Smith 16 sublicens*able 404 sublicense islandable 29 island startable 206 start riffable 29 riff div*able 187 dive appreciatable 28 appreciate layer*able 181 layer dilutable 28 dilute toggl*able 167 toggle proceedable 28 proceed pit*able 100 pit snortable 28 snort categori*able 79 categorise/ ize braggable 27 brag digit*able 52 digitise/ ize debunkable 27 debunk typ*able 51 type sittable 27 sit paus*able 46 pause trekkable 27 trek auction*able 45 auction ageable 26 age hik*able 43 hike encod*able 26 encode sing*along *able 43 sing along piggable 26 pig urbani*able 43 urbanise/ ize toppable 26 top birdie*able 41 birdie wrappable 26 wrap soloable 41 solo boxable 25 box stomach*able 40 stomach chainable 24 chain shovel*able 40 shovel compil*able 24 compile seedable 39 seed diy-able 24 diy optimiz*able 37 optimise/ ize scrunch*ab le 24 scrunch phish*able 37 phish shortable 24 short loop*able 36 loop codifiable 23 codify retirable 35 retire exposable 23 expose ripp*able 35 rip spoofable 23 spoof slappable 35 slap commentable 22 comment socketable 34 socket flashable 22 flash spammable 34 spam How novel are low-frequency words? 17 cach*able 33 cache opt*out*ab le 4 32 opt out Table 4. 66 unlisted forms and their total frequencies in NOW Orthographic variants, marked with asterisks in our dataset, substantially affected frequency counts: 11 Xable items reached 100+ tokens once variants were consolidated. A further limitation concerns our use of the OED as an approximate indicator of lexicalisation. As the OED does not provide a definitive measure of lexical integration, dictionary inclusion cannot be taken as conclusive evidence of lexicalised status. This caveat is especially relevant given the transitional nature of neologisms and the fluid processes through which they become conventionalised: Web 2.0 technologies have led to considerable changes in language usage in that, for example, new word formations may spread and reach a high level of frequency (thus being perceived as widely acceptable by a vast range of speakers) before they have been entered into a dictionary. (Anesa 2019: 4) Beyond delayed lexicalisation for items already in use, the data also revealed cases of re-emerging words previously marked as obsolete. For example, warnable was excluded because it is already attested in the OED. Such instances illustrate the complexity of lexical evolution, in which forms may disappear from common use and later return to circulation. 2.2 Low-frequency listed word 2.2.1 Warnable and OED obsolescence indicators Among the forms excluded due to an existing OED entry, warnable is attested as early as 1449, with only one recorded use marked “obsolete, rare”. In NOW, however, it appears 25 times, mainly in weather reports, especially in the phrase warnable amounts of rain in New Zealand and Australian English. (3) Rainfall values will be a little above normal for the time of year however it will be more settled than the previous week and overall the potential for warnable weather phenomena will be low. (Now Corpus) 4 The form opt-out-able was not excluded because it was not initially recognised as a compound; string irregularities in the corpus prevented its automatic identification as a multi-word base. Chris A. Smith 18 This low-frequency use in the NOW corpus suggests a resurgence of an otherwise obsolete word, showing that lexical lifecycles evolve and that obsolescence can be gradual and reversible (see Smith 2024). The renewed use of warnable is partly explained by its collocational profile and its distribution across varieties (Figure 2) and over time (Figure 3). The term appears almost exclusively in New Zealand sources, in weatherrelated contexts, and frequently co-occurs with amount(s) of rain, as in (4). 5 (4) As of yesterday MetService’s severe weather outlook was moderate confidence of a warnable amount of rain for Buller, Nelson, northern Marlborough, North Taranak […] NZ (17-05-07) Figure 2. Distribution of warnable across varieties in the NOW corpus Figure 3. Chronological distribution of warnable in NOW Warnable meets Kerremans’ (2015) criteria for non-conventionalisation - restricted diffusion, limited discourse range, and low frequency. Its sparse use in NOW also questions its transparency as a deverbal derivative of warn, given the verb’s argument structure. The collocation warnable weather may be influenced by the far more frequent weather warning (24,420 tokens), but the strong concentration of warnable weather in New 5 As rightly pointed out by the editor, this could also mean that the earlier attestation in the OED is unrelated, i.e. that warnable was coined anew in the context of weather reporting, particularly in New Zealand. How novel are low-frequency words? 19 Zealand sources cannot be explained solely by the distribution of weather warning. 2.2.2 Tellable and frequency revival Another OED-attested form, tellable, further illustrates the non-linear nature of lexical evolution. Recorded from 1475 and marked as rare until the 19 th century, it derives from tell, a ditransitive verb (tell [someone] NP [of/ about something] PP). The OED frequency scale shows that tellable has recently resurged, confirming that words may undergo periods of dormancy before re-entering use. Figure 4. Tellable frequency band and chart in OED Despite the OED data, in the NOW corpus tellable has no more than 21 occurrences distributed over a variety of English subcorpora, as shown in Figure 5. Chris A. Smith 20 Figure 5. Occurrences of tellable per country in NOW corpus With greater geographic diffusion than warnable, tellable does not fit the same criteria of non-conventionalisation according to Kerremans (2015). In addition to geographical dispersion, tellable has greater diffusion across contexts. The collocate patterns of tellable show a literal meaning associated with narratives in film, stories, as in (5) and (6): (5) And where it’s hiding is in story - those magical, memorable, tellable moments of truth that have already happened to us - times when the light went on and we connected to a timeless knowing […]. US (15-12-27) Huffington Post (6) Be a ‘tellable’ parent - really listen to what your child is saying. And be an ‘askable’ parent - welcome your child’s curiosity. IE (23-03-28) Irish examiner. Comparing tellable occurrences in (5) and (6) shows that the derived adjective can be used to refer to any of the event participants of the verb tell. In (5) tellable relates to the object / theme of telling (the story) and in (6) tellable relates to the beneficiary of the telling process. To gain further insight into the dataset of low frequency listed words, we examine another exclusion with a surprisingly low frequency. 2.2.3 Damageable and other loans Unlike tellable and warnable, whose low frequencies reflect the argument structure of their base verbs, the rarity of damageable is harder to account for given its transparent derivation from a transitive verb. First attested in 1474 with an obsolete meaning, damageable later developed the modern sense ‘liable to be damaged’, a regular extension of damage and an example of lexical adaptation of a French loan (Durkin 2014; Palmer How novel are low-frequency words? 21 2008, 2015). OED data confirm its very low (sub-0.01) and declining frequency. Figure 6. Frequency of damageable according to the OED This low frequency is all the more surprising and unaccounted for as the verb damage is one of the most frequent words in English with a high frequency band rising until the early 2000s. Figure 7. Frequency of the verb filler damage in the OED Chris A. Smith 22 The low frequency of damageable may be influenced by sociocultural factors (profitability) that affect usage beyond mere linguistic availability. So why is this adjective so rare? One possible explanation is its status as a loanword from French, which may impact its frequency. To explore this, we compare damageable with another borrowing featuring an opaque base or slot filler, Recuperable was excluded from our dataset because it falls below the frequency threshold, with only ten NOW tokens. Despite this rarity, it appears in OED band 3. First attested in 1439, it is treated as a French or Latin loan without a transparent English base. OED data show increased use after 1960, placing it in the 0.01-0.1 pmw range. In contrast, the transparent derivative recoverable is much more frequent (OED band 5), suggesting that transparency supports retention - though this does not explain the unusually low frequency of the equally transparent damageable.The dataset also includes appreciatable, a transparent derivative of appreciate, contrasting with the established borrowing appreciable. According to the OED, appreciable has shifted toward a more figurative sense - ‘capable of being estimated or recognised; of significant extent or degree’ - which may have motivated the creation of the more transparent appreciatable. Figure 8. Recuperable frequency from OED How novel are low-frequency words? 23 2.3 Summary Our objective remains to determine whether this dataset exhibits distinct linguistic properties and to explore what these characteristics reveal about the development of -able adjectives in contemporary English. The results confirm that -able pattern is highly productive for classic transitive verbs, making dictionary attestation an unreliable indicator of lexical establishment. These cases highlight the limitations of dictionary inclusion as a sole measure of conventionalisation. 3. Measuring and tracking the dataset in NOW 2010-2025 3.1 Timeline frequencies Given that the low-frequency status of the items was confirmed in our database, we tracked the chronological development of the 66 unattested forms in NOW to assess their temporal distribution, dispersion, and potential emergence. Raw yearly counts were extracted and converted into relative frequencies based on the size of each annual subcorpus; normalised values were then calculated to control for variation in subcorpus size. Figure 9 shows that normalised frequencies are unevenly distributed across the 15-year period: most forms surface as isolated, low-frequency events, with only a small number displaying sustained presence or shortlived bursts. Higher relative frequencies (lighter shading) cluster in the later years, producing a strongly skewed productivity profile in which a few items rise to local prominence while the majority remain marginal. However, this distribution is distorted by the presence of relatively frequent derivatives that do not meet the threshold for very low frequency. Removing these items - blendable (0.0544), monetizable* (0.0498), cravable* (0.0449), bikable* (0.0337), glancable* (0.0261), sublicensable* (0.0200), divable* (0.0098), layerable* (0.0098), startable (0.0078), and togglable* (0.0077) - sharpens the picture by isolating the genuinely lowand very-low-frequency strata of the dataset. These items display three main tendencies. First, a subset of mediumlow forms - such as printable, shippable, scalable, and tweetable - shows steady, long-term use, though at modest levels compared with higherfrequency derivatives. Second, several adjectives exhibit short-lived bursts of activity: items like snortable or piggable appear briefly in response to topical or domain-specific contexts rather than sustained productivity. Finally, many derivatives are ephemeral coinages, occurring only once or sporadically, suggesting nonce formations or highly restricted contextual usage. Chris A. Smith 24 Figure 9. Heatmap of normalised frequencies for the top 30 of all 66 -able words, 2010-2025. 3.2 Frequency and geographic coverage Following Kerremans (2015)’ criteria for non-conventionalisation - which take into account data from 20 countries with fixed subcorpus sizes - the items with the widest geographic dispersion are also those with the highest overall frequencies. Monetizable, craveable, blendable, and bikeable, for example, occur in 15-19 different country subcorpora, indicating a clear correlation between token frequency and geographic spread. How novel are low-frequency words? 25 Figure 10 presents, for each -able adjective, the number of countries in which it is attested. Figure 10. Number of countries covered by usage (19 to 1) Figure 11. Correlation between timeline frequency and country dispersion monetiz*able crav*able blendable glancable archivable categori*able startable dilutable toggl*able auctionable loopable debunkable compromis… diy-able scrunchable spoofable commenta… spammable cach*able optimiz*able hik*able ageable boxable encod*able stomachable typable soloable proceedable shovelable campable fieldable retirable islandable 0 5 10 15 20 Number of Countries Chris A. Smith 26 Figure 11 plots overall (normalised) frequency against geographic dispersion. The Pearson correlation coefficient is 0.72, indicating a strong positive relationship: items occurring more widely across countries also tend to have higher frequencies over time. This pattern aligns with Kerremans (2015)’s model of neological innovation and conventionalisation, where increasing geographic spread accompanies rising entrenchment. 3.3 Collocate patterns in NOW morphological vs semantic anchoring Collocational patterns provide insights into the stage of conventionalisation of a lexical form. Contextual anchoring plays a crucial role in this process, with genre and text type distribution further influencing the degree of lexical integration. For instance, the adjective stomachable can be assessed through its collocational environment to determine whether it shows morphological anchoring - where the base lexeme recurs in its immediate context - or semantic anchoring, where meaning is reinforced through related conceptual or lexical associations without repetition of the base form. Baayen and Neijt (1997) found that morphological anchoring was more frequent in high-frequency -heid words compared to hapax legomena. In contrast, hapax legomena were more likely to exhibit semantic anchoring, where related words in the discourse provide contextual reinforcement. Their study also revealed a preference for pre-text anchoring rather than post-text (cataphoric) anchoring. Additionally, specialised usage and single-instance occurrences can indicate early-stage conventionalisation. The polysemy potential of Xable adjectives appears to depend partly but not exclusively on the semantic variability of the base word or slot filler. For instance rippable has multiple uses and readings, whereas startable has a single usage in sports discourse. Closely examining Xable patterns can help determine whether low-frequency Xable formations are likely to become integrated into the lexicon or remain marginal. 4. Pattern productivity and pattern extension We consider two types of paradigm extension using our dataset to conduct targeted queries in the NOW corpus: (1) extension to phrasal verbs, both transitive and intransitive, and (2) extension to compound formations. How novel are low-frequency words? 27 4.1 Pattern productivity Neologism formation is often linked to the need to label new concepts - i.e. to fill lexical gaps (Schultink 1961; Štekauer 1998, 2001; Kerremans 2015; Aronoff 2016, 2019; Renner 2020). Yet word formation also fulfils syntactic and pragmatic functions. Many -able formations - such as murderable or ageable - serve discourse-pragmatic purposes (playfulness, evaluativity, branding) rather than naming gaps, as evidenced by the large number of brand-based Xable coinages. Lexical productivity is shaped by more than hapax counts (Palmer 2008; Dalton-Puffer 1996, 2011). Influential factors include family-size effects, base-word frequency, and embedding in phraseological or idiomatic patterns (Baayen 2009). Word class also matters: derived nouns contribute far more to vocabulary expansion than other categories (Plag, Dalton-Puffer & Baayen 1999). As Schmid (2008) argues, nouns have strong reification potential because they designate stable objects, whereas adjectives denote transient properties (Bolinger 1980). Consequently, adjectives may be highly derivable but less stable in the lexicon. Processes of hypostatisation further contribute to the conventionalisation of new forms. Corpus evidence highlights genre variation: -able derivatives are particularly common in technical, marketing, and commercial registers, supporting the view that discourse type shapes semantic development. Polysemy in Xable adjectives often reflects variation in the X-slot, as illustrated by rippable. Methodologically, Baayen and Renouf (1996) emphasise analysing hapax legomena within their wider frequency context rather than in isolation. Examining hapaxes alongside low-frequency neighbours offers a more reliable basis for distinguishing genuine innovations from marginal formations or remnants of declining patterns. 4.2 Pattern of transitive reading The Xable construction is typically understood to denote ‘being suitable for V’, where X conventionally represents a transitive verb - i.e., a verb that can take an argument in the object position. In our dataset, snortable (from the verb snort is attested in 1366 in the primary sense, - and in 1925 slang US for drug taking) and scrunchable (from the verb scrunch, 1825) fit this pattern. As previously observed, ditransitive verbs such as tell and warn can also occupy the X position in the Xable construction and are compatible with passive interpretations for either argument. Chris A. Smith 28 4.2.1 X fillers transitive readings The dataset includes 10 verbs in Table 5 that may deviate from the typical transitive pattern yet serve as fillers for Xable adjectives, with eight forming dictionary-listed Vable derivatives. The definitions provided are glosses we formulated according to the corpus. Word Definition Sittable Suitable for sitting on; also used to describe a player who can be benched in a game or situation. Braggable Impressive enough to be boasted about. Soloable A game or task that can be completed alone; originally attested in a musical and mountaineering context in 1858 (OED). Commentable Open to or allowing comments. Diveable Suitable for diving. Ageable Capable of aging or benefiting from aging. Snortable Suitable for snorting. Campable Suitable for camping; possibly influenced by branding. Glanceable Suitable for glancing. Startable Suitable for starting in the sense given by OED of ‘Of a player: to be in a team’s starting line-up for (a game or match); (also of a coach, manager, etc.) to include (a player) on the team that begins a particular game or match. Also intransitive’. Flashable Flashable (base verb flash 1387; to flash files, i.e. save to a flash drive) in computing, ‘able to be written to or erased from (a flash drive or device)’. Table 5. Transitivity of slot fillers The X slot can thus extend beyond standard transitive fillers, demonstrating the semantic flexibility in the pattern. All of these Xable have idiomatic niche readings: glanceable, flashable and startable exemplify the Idiom Principle (Sinclair 1991, 2004), which posits that language users primarily rely on prefabricated lexical chunks rather than generating meaning through the independent composition of individual words. This principle is illustrated by ageable: Bauer (2018: 182) examined the negative derivative unageable showing that semantic shift (or ‘figurative language’) can be integral to word formation, given that the interpretation of the transitive sense of age shifts from ‘causing to age’ to ‘assigning an age to’. However, in (7), ageable does not have the sense of ‘assigning age’, suggesting that ageable and unageable are not symmetrical, and that unXable has its own pattern. How novel are low-frequency words? 29 (7) …for the sole reason that any wine that is truly ageable from her birth year. NOW (US, 20-02-03) (8) The aim is to make best-of-the-best cabernet sauvignon, ageable for 20 or 30 years. (GB, 22-07-16) Other uses as in (8) from The Times show that ageable has a timeline (10 to 20 years) and corresponds to what is also termed ‘aging potential’. However, there are two uses referring to human beings in the sense ‘being of a certain age’, as in (9) from the Jamaica Gleaner: (9) The man was reversing in the dusk hours. When him was reversing, him knock mi down in the night because him couldn’t see so well. Is a ‘ageable’ man,” McCook related. (CA, 18-10-19) 6 4.2.2 Idiomatic readings and anchoring Certain Xable adjectives require disambiguation because the X slot may correspond to different underlying syntactic structures. As a result, not all tokens instantiate the same morphological pattern. Toppable and socketable illustrate this contrast clearly. Socketable has wide geographic dispersion and growing usage. In (10), multiple tokens of the base word appear in the same text, confirming strong morphological anchoring: (10) Socketables such as Runes can now be overwritten by placing another socketable item into the socket. (SG, 25-01-16) Toppable, by contrast, is concentrated almost exclusively in Irish sports discourse. Example (11) shows its metaphorical reading within Irish football punditry, where it evaluates a qualification group as realistically beatable rather than dominant: (11) …topping an eminently toppable group that features Austria and Wales as the big kahunas. (IE, 20-09-14) Soloable (2012-2024) appears in US, GB, and AU sources, increasing from 35 to 40 occurrences since September 2024. All uses refer to video games that can be completed in single-player mode. Diveable (27 tokens + 2 divable) is attested throughout 2010-2024 and across several varieties. All examples refer to literal diveability of water or depth: 6 The verified source is a Jamaican newspaper however, so the source code should state JM instead of CA in NOW. Chris A. Smith 30 (12)…a fully ‘divable’ and driveable submersible… (AU, 23-06-23) Campable (30 tokens) is almost entirely restricted to New Zealand, where it is primarily a brand name (e.g. an app). Only the earliest example (13) represents a non-brand lexical use: (13) …a carpenter-built campable shell, closed in and with one working bathroom. (US, 10-06-08) Snortable appears 28 times, with a striking concentration in 2017 (16 tokens). Most uses refer to drug administration routes, as in (14): (14) …meow meow is a club drug, generally snortable but also swallowable… (IN, 17-02-17) The earliest use (15) is metaphorical, describing canine behaviour rather than drug intake: (15) …Digby is addicted to the snortable smells in the bushes… (GB, 11-11-14) Snortable shows broad distribution (US, GB, IN, etc.) over 2011-2024. Notably, the analogical form sniffable is recorded in the OED from 1975 in a comparable sense. 4.2.3 X slot fillers and N>V conversion readings The Xable affixation pattern is compatible with an X slot that can be filled by nouns as well as verbs. Many of the adjectives in the dataset correspond to multi-categorial words, and specifically intransitive verbs which originate from earlier nouns. All of the following Xable adjectives contain slot fillers originating diachronically as nouns. We provide glosses based on meanings observed in the corpus. Word Noun Definition toppable top (oe) Can be placed on top of something; derived from top (v.), attested since 1300. ageable age (1275) Suitable for aging, benefiting from aging, v. attested 1440 riffable riff (1933) Suitable for improvisation or playing riffs; from riff (v.), 1935. socketable socket (1330) Capable of fitting into a socket; from socket (v.), attested since 1533. How novel are low-frequency words? 31 stomachable stomach (1340) Tolerable; derived from stomach (v.), attested since 1523. islandable island (oe) Capable of being isolated or made into an island; derived from island (v.), attested since 1661) auctionable auction (1595) Suitable for auctioning; from auction (v.), attested since 1884. chainable chain (1300) Able to be linked or connected in a sequence, from chain v attested since 1377 boxable box (oe) Suitable for boxing, from box v 1450 campable camp (1525) Suitable for camping/ setting up camp, from base verb to camp 1543 ‘to sojourn in a tent’ Table 6. Xable fillers (16) The CPA said 85,959 metric tonnes of auctionable and destroyable goods (about 8,000 TUEs). (BD, 19-04-26) All Xables in Table 6 are compatible with transitive readings which means that the potential processual reading is attributed to the event participants. 4.2.4 Hapax legomena in NOW: needle in a haystack The rarest forms - true single-occurrence items - are often surprisingly difficult to locate in a large corpus, representing the proverbial ‘needle in a haystack’. Apart from tinsurable (an erroneous hit), most hapax -able forms required manual verification, since very rare items may reflect noise, misspelling, or nonce formation. Collocate-based inspection nonetheless enabled us to identify several genuine singletons alongside additional low-frequency forms. Some hapaxes turn out to be orthographic errors. Quoffable appears once in The Guardian (17), but closer inspection shows it is a misspelling of quaffable, which has 410 occurrences in NOW: (17) I’m expecting this to be very quoffable, hence the large quantities. (GB, 20-01-31) Similarly, cachetable appears once (18), almost certainly for cacheable: (18) …takes up space within the PerconaFT cachetable just like any other PerconaFT dictionary. (IN, 16-09-28) Chris A. Smith 32 Other items illustrate how low counts may mask orthographic or semantic variation. Mouthable has nine occurrences, mostly in the sense ‘suitable for putting in the mouth’, as in (19), though occasional metaphorical uses also occur, as in (20): (19) … any “mouthable” toy and toys for children under age 3. (US, 22-01-04) (20) humour-mongering, unexpected props, mouthable stories, human heads… (IN, 16-11-19) Hoverable has six hits but only three distinct occurrences; in (21), the base verb hover appears in the preceding context, serving as a clear morphological anchor: (21) …several boxouts, themselves packed with hoverable stuff. (AU, 17-07-15) Two further fillers in our dataset include prepositional structures: singalongable (23 tokens) and opt-outable (32 tokens). Opt-outable is especially skewed, with ten Canadian occurrences in 2011 and eight in 2016 from a single source: (22) …the fee’s opt-outable nature became a subject of debate. (CA, 16-11-07) These cases show why methodological refinement is essential: distinguishing genuine rare formations from noise or variation requires targeted verification rather than exhaustive dataset trawling. To pursue this systematically, we therefore focus subsequent corpus queries on Xable formations with unconventional verb-particle fillers, where productive innovation is most likely to occur. 4.3 Unconventional V particle patterns Verb-particle derivatives in -able are often treated as marginal or problematic for rule-based morphology (Chapman 2008; Farrell 2005; Miller 2014; Bagasheva 2024). Miller (2014), for instance, notes that V + particle + -able formations (e.g. wade-through-able, explain-away-able) are acceptable, whereas double derivations such as *explainable-awayable are ruled out, in contrast with the productive -er pattern (picker-upper). Farrell (2005: 103) similarly observes that VP lexemes participate in affixation and that (un-)V-P-able complexes are not uncommon online. How novel are low-frequency words? 33 To explore these extensions systematically, we conducted targeted searches in the NOW corpus using strings such as *up*able, *out*able, *off*able, *through*able, *without*able and *away*able, followed by manual filtering. Overly noisy patterns (e.g. over*able, at*able) were discarded. The goal was not to exhaustively list all coinages, but to identify recurrent structural types and assess how far Xable extends beyond canonical transitive bases. Table 7 summarises the main filler types attested in the low-frequency dataset. Rather than being isolated curiosities, these types cluster around a small set of recurrent patterns, which we briefly illustrate below. Filler type Example V (transitive) + -able donatable V (intransitive) + -able hoverable, runawayable V, N or NP + -able tax-writeoff-able, off-road-able, word-of-mouthable V (transitive + particle) + -able opt-outable, ungetthroughable V (transitive + two particles) + -able make-out-with-able V + coordinated particles + -able dip-in-and-out-able, hop-on-and-offable (V)able + (particle)able (double affixation) unliveable-withoutable Compound verb + -able mouth-poppable, car-toppable Phrasal with conjugation + -able can’t-ride-without-able V the unVable get through the un-get-throughable, outwork the outworkable Table 7. Overview of low-frequency V-able structures in NOW 4.3.1 *UP* able Many items follow the straightforward pattern V + particle + -able, with X interpreted as a processual verb phrase. Several UP-able formations illustrate this use: (23) He returned with The Thin Red Line, a lush, star-studded war movie that managed to make you think “Wow”, even when the shut-uppable voiceover became too airy-fairy and the shots of parrots became too ponderous. (IE, 16-05-05) (24) The house in question […] would be […] very dry, lock-up-able and stood up in bad weather. (AU, 23-06-08) Chris A. Smith 34 (25) ‘Would be good for babies new to pick-uppable solid food’, they said. (US, 22-09-30) In all three cases (23) to (25), the derivative does not fill a lexical gap (e.g. tolerable, jailable already exist), but provides an expressive, often humorous paraphrase of “can be shut up / locked up / picked up”. A two-particle base also surfaces: (26) Still, those cons could be put-up-withable for $900 if all you really care about is excellent performance. (US, 24-11-18) Here, put up with is fully integrated as the X-slot filler, showing that speakers treat multiword VPs as legitimate morphological bases. Comparable patterns arise with OUT and OFF, where the X slot may contain a phrasal verb or an NP reanalysed as a verbal base, for instance, chuck-out-able (“all chuck-out-able afterwards”), off-road-able (“surprisingly off-road-able”), runawayable (“Data is not runawayable”), extending the intransitive run away to a transitive-like meaning (‘capable of being run away with / removed’). Finally, the query also revealed creative uses of proper names as Vparticle-like bases: (27) As part of Groupon’s strategy […] to capture more of the 80-plus Grouponable moments happening every year […]. (US, 21-11- 02) Here, Grouponable functions as an affixed brand-based analogue to VPable forms, reinforcing that such coinages operate at the expressive end of the derivational space rather than filling lexical gaps. 4.3.2 *OUT* able A search for outable returned over 200 raw hits, requiring manual sorting to isolate genuine V-particle-able formations. Both OUT V able and V OUT able patterns appear, with several hapax formations suggesting an emergent productive extension of the schema. The most frequent simple VPbased types include: (28) And a foraging walk in nature yields treasures […]. All free, all chuck-out-able afterwards. (AU, 22-10-06) (29) …still perfectly chuck-aboutable when you want to dodge holes and bumps on the track surface… (GB, 22-10-27) How novel are low-frequency words? 35 Here, chuck out/ about is transparently reanalysed as the verbal base, giving a processual reading consistent with the Xable template. A more structurally complex instance is the coordinated-particle base in: (30) …a spry and very dip-in-and-out-able anthology… (GB, 25-01- 08) Further extension is illustrated by figureoutable, attested 14 times (24 with spelling variants), lexicalised through the title of a bestselling book: (31) Everything Is Figureoutable by Marie Forleo. (GB, 22-06-08) Other creative VP-like bases include compound or serial verbs, such as: (32) The mouth-poppable frozen corn dogs are now available nationwide… (US, 23-02-15) and adverbially modified uses: (33) …making it an easily storable or easily leave-out-able home decor statement. (US, 23-05-25) Multi-particle VPs are also attested, showing that speakers readily treat phrasal sequences as morphological units: (34) …to help make us appear at least semi-attractive and make-outwith-able. (US, 11-01-06) The construction also supports analogical extensions such as <V the un- Vable>: (35) He outworks the outworkable. (U, 23-12-19) and modal-like uses: (36) …is it outlaw-able? (US, 12-09-07) A striking set of forms involves NP fillers reanalysed as verbal bases, e.g. the noun phrase word of mouth in: (37) …features […] should be ‘word-of-mouthable’. (US, 10-03-29) Chris A. Smith 36 Brand-based X-slot fillers also appear, parallel to VP-able coinages: (38) She advocates helping people […] get out and about more easily. (AU, 18-10-23) - Getaboutable (39) …getting the cool, exclusive YouTubable footage. (GB, 17-02- 25) These items confirm that the schema extends to proper names that behave like quasi-verbal units within the construction. Taken together, these forms show that -able readily attaches to VPs, coordinated particles, NP/ compound bases, and brand names, and that speakers use Xable to impose transitivity or processual meaning on nonverbal material. The pattern supports conversion and metonymic reinterpretation (Bauer 2018), and clearly operates at the expressive, innovative end of the derivational system. 4.3.3 *OFF*able The search for offable yielded fewer cases than outable, but the attested items show clear parallels with the VP-based extensions described above. Several forms instantiate a [V + coordinated particle] + able or [NP/ AP + off] + able pattern. A coordinated-particle base appears in: (40) …a reliable, efficient, affordable and truly democratic hop-onand-off-able transport system? (GB, 17-12-10) Here hop on/ hop off - a fixed phrasal-verb pair used in public-transport contexts - is incorporated as a unified X-slot filler. Other formations show X as an NP or compound reanalysed as a processual base: (41) I’ve been making modest, tax-write-off-able […] donations since the early 1980s. (CA, 13-03-10) (42) It’s surprisingly off-road-able! It can climb over light blankets and other thin things on the ground. (US, 20-06-17) Both tax write-off and off road behave here as verbalised compounds, confirming that Xable can attach to nominal or adjectival constituents that speakers reinterpret as processes. How novel are low-frequency words? 37 Negative formations also surface, typically without attested base forms, showing that unXable may arise independently rather than through prefixation of an existing Xable: (43) …the woman with the un-shake-off-able liking for the word burr… (GB, 10-08-08) (44) …their un-switch-off-able campaign last year. (NZ, 17-08-14) In both cases, no corresponding shake-offable or switch-offable is attested, suggesting that speakers build the negative form analogically on the VP rather than on an existing adjective. A further extension involves the reinterpretation of normally intransitive V+particle sequences as transitive-like Xable bases: (45) …which suggests Casemiro’s earlier knock is not run-off-able. (SG, 18-11-11) Here run off (intransitive) is reanalysed as something that can be “run off” (i.e. worked off, recovered from), illustrating how the construction licenses coercion of argument structure. These examples, drawn from a geographically diverse set of sources, reinforce that the OFF-able pattern - though less frequent - participates in the same family of VP-, NP-, and compound-based extensions observed elsewhere in the dataset. 4.3.4 *WITHOUT*able: two special extensions The withoutable query in NOW yielded two atypical formations not attested elsewhere in the dataset. The first is a rare case of double -able affixation: (46) …she has […] become frighteningly unliveable-withoutable. (IE, 19-03-13) This appears to be the only double -able derivative in the corpus (see Chapman 2008; Miller 2014: 100; Bagasheva 2024). Its structure is plausibly analogical to the rhyming double-affix patterns (-er/ -er, -ee/ -ee) found in V-P lexemes (Farrel 2005: 103). The second item incorporates a conjugated verb phrase into the X-slot: (47) What makes it so useful/ special/ can’t-ride-without-able? (GB, 18-09-29) Chris A. Smith 38 Here X is filled by can’t ride without, producing a syntactic hybrid in which the modal auxiliary remains embedded within the morphological base. In context, the formation echoes entrenched expressions such as can’t live without, showing that speakers can package whole phrasal sequences into an -able adjective. These two forms represent a marked extension of the Xable schema to multiword, syntactically complex fillers. Both are transparent in context, indicating that such innovations remain interpretable and do not exceed the productive limits of the pattern. 4.3.5 *AWAY*able The query awayable returned the largest set of VP-based formations, including 15 tokens of runawayable and several related items (put-awayable, un-click-away-able, un-look-away-able, semi-throwawayable). These follow the same VP-integration patterns observed earlier. (48) Data is not runawayable. But made inaccessible. (PH, 19-01-23) Runawayable is notable because it coerces the normally intransitive run away into a transitive-like reading (‘capable of being run away with / removed’) applied to an inanimate subject (data). (49) This has to be shareable, semi-throwawayable stuff. (IE, 18-11- 30) Here semimodifies throw-awayable, extending an already processual V+P base. (50) The show […] has become an un-look-away-able train wreck. (US, 24-02-21) Un-look-away-able compresses the idiomatic expression can’t look away from into a single adjective, showing that -able can incorporate even figurative, multiword verbal bases. 4.3.6 *THROUGH*able The query throughable yielded several extensions, most of them incorporating complex VP structures. (51) …getting Black people through the un-get-throughable. (US, 22-06-15) How novel are low-frequency words? 39 This example follows the pattern <V the unVable>, previously seen in work the unworkable, now applied to get through, producing an intensified, emphatic form. Another formation highlights metalinguistic distancing via quotation marks: (52) …a hairdo deemed ‘un-see-through-able’ by the airport scanners. (US, 14-07-05) Here the agent complement (by the airport scanners) reinforces the processual interpretation of see through. Two tokens of wriggle-through-able were also found (no available context). Finally: (53) …ways that seem so plainly fixable, fight-through-able. (CA, 17-05-05) Fight-through-able appears alongside fixable, stressing Xable’s discourseorganising role in evaluative sequencing. 4.4 Overview of observed pattern extensions Our initial search for low-frequency -able forms in NOW generated numerous compound structures, which were excluded from the main dataset but nonetheless recur among the 66 unattested items. One clear case is car-toppable: (54) The boat is car-toppable by one person… (IE, 16-10-11) Its interpretation requires the compound verb car-top ([[car top]V + able]), not toppable modified by car. As in machine-readable, omitting the nominal element would remove essential meaning. A second example is coffee-dunkable: (55) …tempting, coffee-dunkable pastries… (US, 22-01-21) Here coffeeis a weak modifier; the core base is the independently attested dunkable (95 NOW tokens). Particle-based queries also produced rare analogical formations such as can’t-ride-without-able and unliveable-withoutable, which confirm that able attaches not only to compounds but also to multiword or syntactically complex bases. As Farrell (2005: 103) notes, such derivatives support a compound-like analysis: derivational affixes target whole lexical units, regardless of internal complexity. Chris A. Smith 40 Across the unattested items, -able extends to: phrasal verbs (opt-outable, ungetthroughable), coordinated particles (dip-in-and-out-able, hop-on-and-offable), - NP/ AP fillers (tax-write-off-able, off-road-able), brand names (Grouponable, YouTubable), compound verbs (mouth-poppable, car-toppable), double affixation (unliveable-withoutable), conjugated phrasal bases (can’t-ride-without-able), - V the unVable forms (get through the un-get-throughable). These patterns show that -able productively licenses transitivity, coercion, and metonymic extension, enabling highly creative X-slot realisations well beyond canonical verb stems. 5. Discussion and outlook The goal of the study was to test a corpus-driven procedure for identifying emerging lexical items by surveying low-frequency Xable forms and tracing their micro-diachronic behaviour over 15 years. Although insufficient for gauging conventionalisation, the method successfully identified several extensions of the Xable schema. The behaviour of low-frequency tokens confirms the type-productivity of Xable. Across the dataset, -able continues to license new formations from a broad set of slot fillers - transitive, intransitive, prepositional and phrasal verbs, compounds (tax-write-off-able), and even phrasal elements (can’t-ride-without-able). These forms are consistently interpreted as processual, implying susceptibility to an event (e.g., breakable). Very lowfrequency forms, especially those predicted through particle-based searches, yield innovative VP-, NP-, and compound-based extensions that increase the semantic coverage of the schema (Goldberg 2019). The data raise several theoretical questions: 1. Derivational vs. inflection-like behaviour. Xable is a derivational suffix, yet many formations behave like VP-based units and display coerced or newly assigned transitivity. How should such inflection-like behaviour be modelled? 2. Potential suffixoid status. Several patterns suggest a drift toward suffixoid behaviour (Booij & Hüning 2014), strengthened by analogical pull from adjectives such as able, suitable, capable, available and their compounds, and reinforced by near-synonymy with many Xable forms. 3. Parallel with -like. Xable shows close parallels with -like, another adjective-derived affixoid that attaches to complex bases and allows ex- How novel are low-frequency words? 41 pressive, coerced readings. Does the identity of able as a free adjective facilitate the same kind of expansion? 4. Phrasal Xable and hybrid affixoids. The emergence of uppable, outable, offable, awayable, throughable and their VP-based derivatives suggests that Xable may undergo phrasal reanalysis, supporting an analysis in terms of hybrid affixoid behaviour rather than strictly word-based derivation. Our findings highlight the limits of frequency-based diagnostics. As Jackendoff and Audring (2016: 479) note for Xed out, frequency alone does not adequately capture schema productivity. Productivity also depends on factors such as chronological dispersion, register, and discourse domain (Bauer 2001, 2005). Complementary perspectives emphasise interdisciplinary, data-driven approaches (Baayen 2009) and the need to combine quantitative and qualitative evidence (Fernández-Domínguez 2013; see also Smith 2023 on metapragmatic productivity). The present data further show that low productivity is gradient rather than confined to hapaxes. This supports a model of morphological productivity as part of a dynamic, hierarchical lexical ecosystem shaped by semantic, historical, stylistic, and cognitive factors (Baayen 2009; Fernández-Domínguez 2013). Within this broader framework, Goldberg’s (2019) concept of semantic coverage provides a useful lens for understanding how Xable patterns expand and reorganise across the lexicon. References Anesa, P. (2019). Lexical Innovation in World Englishes: Cross-Fertilization and Evolving Paradigms. London: Routledge. Aronoff, M. (1976). Word Formation in Generative Grammar. Cambridge, MA: MIT Press. Aronoff, M. (2016). Competition and the lexicon. In: A. Elia, C. Iacobini & M. Voghera (Eds.). Livelli di Analisi e Fenomeni di Interfaccia. Rome: Bulzoni. 39- 52. Aronoff, M. (2019). Competitors and alternants in linguistic morphology. In: F. Rainer, F. Gardani, W. U. Dressler & H. C. Luschützky (Eds.). Competition in Inflection and Word-Formation. Dordrecht: Springer. 39-66. Aronoff, M. & Anshen, F. (2001). Morphology and the lexicon: lexicalization and productivity. In: A. Spencer & A. M. Zwicky (Eds.). The Handbook of Morphology. Oxford: Blackwell. 237-247. Arndt-Lappe, S. (2014). Analogy in suffix rivalry: the case of English -ity and ness. English Language and Linguistics 18 (3): 497-548. Baayen, R. H. (1992). Quantitative aspects of morphological productivity. In: G. Booij & J. van Marle (Eds.). Yearbook of Morphology 1991. Dordrecht: Kluwer. 109-149. Chris A. Smith 42 Baayen, R. H. (1993). On frequency, transparency, and productivity. In: G. Booij & J. van Marle (Eds.). Yearbook of Morphology 1992. Dordrecht: Kluwer. 181- 208. Baayen, R. H. (2009). Corpus linguistics in morphology: morphological productivity. In: A. Lüdeling & M. Kytö (Eds.). Corpus Linguistics: An International Handbook. Vol. 2. Berlin: Mouton de Gruyter. 899-919. Baayen, R. H. & Lieber, R. (1991). Productivity and English derivation: a corpusbased study. Linguistics 29: 801-844. Baayen, R. H. & Neijt, A. (1997). Productivity in context: a case study of a Dutch suffix. Linguistics 35: 565-587. Baayen, R. H. & Renouf, A. (1996). Chronicling the times: productive lexical innovations in an English newspaper. Language 72: 69-96. https: / / doi.org/ 10.2307/ 416794 Bagasheva, A. (2024). An upper take on doubler-uppers. Languages 9 (3): 91. https: / / doi.org/ 10.3390/ languages9030091 Bauer, L. (1983). English Word-Formation. Cambridge: Cambridge University Press. Bauer, L. (2001/ 2004). Morphological Productivity. Cambridge: Cambridge University Press. Bauer, L. (2005). Productivity: theories. In: P. Štekauer & R. Lieber (Eds.). Handbook of Word-Formation. Dordrecht: Springer. 315-334. Bauer, L. (2006). Competition in English word formation. In: A. van Kemenade & B. Los (Eds.). The Handbook of the History of English. Oxford: Blackwell. 177- 198. Bauer, L. (2018). Conversion as metonymy. Word Structure 11 (2): 175-184. Bauer, L., Lieber, R. & Plag, I. (2013). The Oxford Reference Guide to English Morphology. Oxford: Oxford University Press. Bolinger, D. L. (1980). Language: The Loaded Weapon. London & New York: Longman. Booij, G. (2010). Construction Morphology. Oxford: Oxford University Press. Booij, G. (Ed.). (2018). The Construction of Words: Advances in Construction Morphology. Cham: Springer. Booij, G. & Masini, F. (2018). Partial motivation, multiple motivation: the role of output schemas in morphology. In: G. Booij (Ed.). The Construction of Words: Advances in Construction Morphology. Cham: Springer. 59-80. Booij, G. & Hüning, M. (2014). Affixoids and constructional idioms. In: R. Boogaart, T. Colleman & G. Rutten (Eds.). Extending the Scope of Construction Grammar. Berlin: De Gruyter Mouton. 77-106. Bybee, J. (2010). Language, Usage and Cognition. Cambridge: Cambridge University Press. Chapman, D. (2008). Fixer-uppers and passers-by: nominalization of verb-particle constructions. In: S. M. Fitzmaurice & D. Minkova (Eds.). Studies in the History of the English Language IV. Berlin: Mouton de Gruyter. 265-299. Dalton-Puffer, C. (1996). The French Influence on Middle English Morphology. Berlin: De Gruyter Mouton. https: / / doi.org/ 10.1515/ 9783110822113 Dalton-Puffer, C. (2011). Productive or non-productive? the Romance element in Middle English derivation. In: F. Fernández, M. Fuster Márquez & J. J. Calvo (Eds.). English Historical Linguistics 1992. Amsterdam: John Benjamins. 247- 260. Davies, M. (2016-). Corpus of News on the Web (NOW). https: / / www.englishcorpora.org/ now/ How novel are low-frequency words? 43 Durkin, P. (2014). Borrowed Words: A History of Loanwords in English. Oxford: Oxford University Press. Fischer, R. (1998). Lexical Change in Present-Day English. Tübingen: Narr. Goldberg, A. E. (2019). Explain Me This: Creativity, Competition, and the Partial Productivity of Constructions. Princeton: Princeton University Press. Hohenhaus, P. (2005). Lexicalization and institutionalization. In: P. Štekauer & R. Lieber (Eds.). Handbook of Word-Formation. Dordrecht: Springer. 353-373. Huyghe, R. & Varvara, R. (2023). Affix rivalry: theoretical and methodological challenges. Word Structure 16 (1): 1-23. Jackendoff, R. (1997). The Architecture of the Language Faculty. Cambridge, MA: MIT Press. Jackendoff, R. & Audring, J. (2016). Morphological schemas. The Mental Lexicon 11 (3): 467-493. Kerremans, D. (2015). A Web of New Words. Frankfurt am Main: Peter Lang. Lipka, L. (1977). Lexikalisierung, idiomatisierung und hypostasierung als probleme einer synchronischen wortbildungslehre. In: H. E. Brekle & D. Kastovsky (Eds.). Perspektiven der Wortbildungsforschung. Bonn: Bouvier. 155-164. Lipka, L. (2002). English Lexicology: Lexical Structure, Word Semantics and Word- Formation. Tübingen: Narr. Mattiello, E. (2017). Analogy in Word-Formation: A Study of English Neologisms and Occasionalisms. Berlin: De Gruyter Mouton. Oltra-Massuet, I. (2014). Deverbal Adjectives at the Interface. Berlin: Mouton de Gruyter. Oxford English Dictionary. (n.d.). The Oxford English Dictionary. Oxford: Oxford University Press. Palmer, C. C. (2008). Borrowed derivational morphology in Late Middle English. In: S. M. Fitzmaurice & D. Minkova (Eds.). Studies in the History of the English Language IV. Berlin: Mouton de Gruyter. 231-264. Palmer, C. C. (2015). Measuring productivity diachronically. English Language and Linguistics 19 (1): 107-129. Plag, I. (1999). Morphological Productivity. Berlin: Mouton de Gruyter. Plag, I. (2003). Word-Formation in English. Cambridge: Cambridge University Press. Plag, I., Dalton-Puffer, C. & Baayen, R. H. (1999). Morphological productivity across speech and writing. English Language and Linguistics 3 (2): 209-228. Renner, V. (2020). An ecosystem view of word-formation. The Mental Lexicon 15 (1): 4-20. Schmid, H.-J. (2008). New words in the mind: concept-formation and entrenchment of neologisms. Anglia 126 (1): 1-36. Schmid, H.-J. (2011). English Morphology and Word-Formation: An Introduction. Berlin: Erich Schmidt Verlag. Schmid, H.-J. (2014). Lexico-grammatical patterns, pragmatic associations and discourse frequency. In: T. Herbst, H.-J. Schmid & S. Faulhaber (Eds.). Constructions, Collocations and Patterns. Berlin: Mouton de Gruyter. 239-293. Schmid, H.-J. (2015). The scope of word-formation research. In: P. O. Müller, I. Ohnheiser & S. Olsen (Eds.). Word-Formation: An International Handbook of the Languages of Europe. Berlin: De Gruyter. 1-21. Schmid, H.-J. (2020). The Dynamics of the Linguistic System. Oxford: Oxford University Press. Schultink, H. (1961). Produktiviteit als morfologisch fenomeen. Forum der Letteren 2: 110-125. Chris A. Smith 44 Sinclair, J. (1991). Corpus, Concordance, Collocation. Oxford: Oxford University Press. Sinclair, J. (2004). Trust the Text: Language, Corpus and Discourse. London: Routledge. Smith, C. A. (2020). A case study of -some and -able derivatives in the OED3. Lexis 16. https: / / doi.org/ 10.4000/ lexis.4793 Smith, C. A. (2023). Productivity from a metapragmatic perspective. Languages 8 (2): 92. https: / / doi.org/ 10.3390/ languages8020092 Smith, C. A. (2024). Assessing lexicographic obsolescence and historical frequency indicators. In: J. Martín Arista & A. E. Ojanguren López (Eds.). Structuring Lexical Data and Digitising Dictionaries. Leiden: Brill. 356-393. Štekauer, P. (1998). An Onomasiological Theory of Word-Formation. Amsterdam: John Benjamins. Štekauer, P. (2001). Fundamental principles of an onomasiological theory of English word-formation. Onomasiology Online 2: 1-42. Chris A. Smith Université Caen Normandie, CRISCO UR 4255