Languages differ with respect to whether measure phrases (MPs) can co-occur with bare relative gradable adjectives (GAs), and if so, what interpretations arise.1 In languages like English, some relative GAs can co-occur with an MP in their bare forms, and the MP receives an absolute interpretation. In (1a), the MP 8m measures the height of the pine tree itself, not in comparison with another entity. Other relative GAs, such as heavy, cannot co-occur with an MP in the absence of an overt comparative morpheme (1b).

English shares this phenomenon with languages such as German, Dutch, Italian, and French, although the specific lexical items that can or cannot co-occur with MPs in their bare forms vary even between closely related languages (Schwarzschild 2005, Beck et al. 2009).

Then there are languages that never allow MPs to co-occur with bare relative GAs. For example, Spanish and Russian are reported to be of this type (Schwarzschild 2005, Beck et al. 2009, Sawada & Grano 2011).2

The third pattern, exemplified by Japanese, is that MPs co-occurring with relative GAs consistently receive differential interpretations (Snyder et al. 1995, Beck et al. 2004, Oda 2008, Beck et al. 2009, Hayashishita 2009, Kubota 2011, Sawada & Grano 2011, Watanabe 2013, Aonuki 2024c). In (2), the MP 8m measures how much taller the pine tree is than another salient entity, in this case the maple tree.

To my knowledge, Japanese and Korean are the only languages that have been reported to be of this type.3^,4

Gitksan patterns with Japanese with respect to MPs.5 In (3), which involves a relative GA ’wii ’nakw ‘long’, even though there is no overt comparative morpheme or overt standard, the MP k’i’y=hl t’imk’aax ‘one arm span’ receives a differential interpretation, measuring how much longer the table is than another table.6

On the other hand, MPs receive an absolute reading when a co-occurring bare GA is minimum-standard.7 In (4), with a minimum-standard GA k’ak ‘open’, the MP k’i’y=hl hlek moos ‘one inch’ receives an absolute interpretation, measuring how wide the door is open.8

In Gitksan, the contrast between relative and minimum-standard GAs extends to constructions without MPs. Relative GAs can receive comparative readings even without an MP (5).9

The overt standard is optional in (5), meaning that the literal translation of Michael is tall can receive a comparative reading, ‘Michael is taller.’ On the other hand, bare minimum-standard GAs cannot receive a comparative reading, with or without an overt standard phrase (6).

Gitksan differs from Japanese in this regard. In Japanese, the literal translation of The pine tree is tall, a version of (2) without the MP, does not have a comparative reading; it simply has a positive reading, and a comparative reading requires an additional morpheme (see Section 10).10 Gitksan also differs from Japanese in having an overt, morphologically free morpheme, k’aa, which gives rise to comparative readings even with minimum-standard GAs.

In summary, Gitksan shows two contrasts between relative and minimum-standard GAs that need to be explained. With respect to MPs, bare relative GAs give rise to differential interpretations, while bare minimum-standard GAs give rise to absolute interpretations. Without MPs, bare relative GAs can receive comparative interpretations, while minimum-standard GAs cannot. I will argue that these two contrasts between relative and minimum-standard GAs receive a unified explanation if the context-sensitivity of relative GAs is built into their lexical denotations. My analysis builds on Oda’s (2008) analysis of differential MPs in Japanese, with some modifications for how the value of a context-sensitive standard degree is determined. In fact, the pattern in Gitksan is what is predicted by the inherently context-sensitive denotations of GAs. Thus Gitksan data both provide support for this approach and raise a question why other languages that have received similar analyses (Oda 2008 on Japanese, Breakstone 2012, Cariani et al. 2023, 2024, Wellwood 2024 on English) do not exhibit the predicted patterns. I will also argue that the presence of the morpheme k’aa makes implausible an alternative account of the differential MP readings involving a covert coercion operator for Gitksan, which has been proposed for Japanese by Sawada & Grano (2011) as a repair mechanism available due to the absence of an overt and free comparative morpheme in Japanese.

The remainder of the paper is structured as follows. Section 2 provides the necessary background on degree semantics, with a focus on scale structures (2.1), where the context-sensitivity of GAs is located in the grammar (2.2), and different strategies by which natural language achieves comparison (2.3). After providing some background on Gitksan (Section 3), Section 4 presents Gitksan data that demonstrate the scale-structure differences in MP interpretations (4.1) and the availability of comparative readings without a comparative morpheme (4.2). An analysis of the Gitksan data is developed in Section 5. After reviewing Oda’s analysis for Japanese (5.1), I argue for similar, inherently context-sensitive denotations of GAs (5.2), and provide an account of the scale-structure difference in MP interpretations (5.3). I then make a slight revision to the GA denotations to better predict when positive vs. comparative readings arise and spell out my account of scale-structure differences in availability of comparatives (5.4, 5.5). I will demonstrate that the analysis makes a correct prediction about another property of comparative readings without a comparative operator in Gitksan, namely compatibility with what are called ‘crisp judgment’ (Kennedy 2007a) contexts. In Section 6, I consider four possible alternative analyses and explain why they do not work for Gitksan. The latter half of the paper is dedicated to further descriptions of comparative and superlative constructions in Gitksan. Specifically, I investigate the semantics of the morpheme k’aa, which achieves comparative and superlative readings even with minimum-standard GAs (Section 7), and propose to analyze it as a superlative operator (Section 8). I also argue that the value of a context-sensitive standard degree of a relative GA is narrowed down by consideration of alternatives to the subject (Section 9). Finally, in Section 10, I revisit Japanese with respect to predictions of the inherently context-sensitive GA denotations (Oda 2008) and discuss what Gitksan tells us about this approach more broadly and the theoretical and empirical typologies of degree semantics and comparatives. Section 11 concludes.

Gitksan is a Tsimshianic language spoken in northern British Columbia, Canada. It constitutes the Interior Tsimshianic branch along with a neighbouring language, Nisga’a. There are approximately 255 fluent speakers as of 2022 (Gessner et al. 2022).

Unless otherwise noted, Gitksan data presented in this paper are from my fieldwork with two speakers, Vincent Gogag (VG) and Hector Hill (HH). Data were elicited using the standard methodologies in semantic fieldwork (Matthewson 2004, Bochnak & Matthewson 2015) including translation and acceptability judgment tasks, with discourse contexts established verbally and/or with visual aids.

This paper presents the first systematic documentation of and formal work on degree constructions in the Tsimshianic language family. Apart from preliminary work in Aonuki (2023) and Aonuki (2024b), Bicevskis et al. (2017) describe amount comparatives and equatives (346-7); Rigsby’s (1986) grammar includes degree questions (95-96); and Tarpent’s (1987) grammar of Nisga’a documents gradable nominals (244-6) and some comparative sentences (232, 306). See also Forbes (2012) for detailed discussions of adjectives as a syntactic category in Gitksan.

The word order in Gitksan is VSO. NPs are preceded by what are known in the Tsimshianic literature as connectives, =hl, =s∼ t17 and dip (Rigsby 1986, Tarpent 1987). Apart from the fact that =hl precedes common nouns while =s∼t precedes singular proper names and dip precedes plural or coordinated proper names, connectives are not sensitive to any other semantic properties, such as definiteness and specificity (Davis 2018).

An example of a positive construction is provided in (26) (see also fn. 6).

Some adjectives have a distinct plural form. For example, hi’niiluxw is a plural form of ’nakw.

In Section 5, I demonstrated that postulating inherently context-sensitive and comparative denotations of GAs similar to Oda’s (2008) analysis of Japanese can provide a unified explanation for why, in the absence of an overt comparative morpheme, comparative interpretations and differential MP interpretations are available with a relative GA but not with a minimum-standard GA. In this section, I discuss four possible alternative analyses and why they do not account for the Gitksan data.

It was demonstrated in Section 4 that minimum-standard GAs cannot receive a comparative interpretation on their own, regardless of the presence of a standard phrase (see (80), repeated from (62), and (81), repeated from (39)).

How is a comparative reading achieved with these predicates, then? As it turns out, Gitksan has an overt degree morpheme k’aa, which can give rise to a comparative reading regardless of the scale structure of the co-occurring GA. Adding k’aa to (80) and (81) make them acceptable with a comparative reading ((82), (83)).

The compatibility with minimum-standard GAs is evidence that k’aa is an ‘explicit’ degree operator akin to -er rather than a morpheme that relies on the context-sensitivity of relative GAs to achieve comparison, like pos (Kennedy 2007a; see Section 2.3). Consistent with this view, k’aa is also compatible with relative GAs (84) and acceptable in crisp judgment contexts (85) (see Section 2.3, 5.6).

Regardless of the scale structure of the GA, the standard phrase remains optional in the presence of k’aa; this is tested in (83) and (84).

In the descriptive literature, k’aa is glossed as ‘exceedingly’ in Rigsby’s (1986) grammar of Gitksan (155). Tarpent’s (1987) grammar of a neighbouring language, Nisga’a, glosses k’aa as ‘most, excessively, extremely’ and describes that it contributes meanings of intensification, comparatives, or superlatives (389-391). Indeed, k’aa appears in translations of superlatives as well. Just like with comparatives, k’aa is obligatory if the GA is minimum-standard (as in (86) and (87)), and optional if the GA is relative (88).

Finally, given the descriptions of k’aa by Rigsby (1986) and Tarpent (1987) above, it should be tested whether k’aa can be used as an intensifier akin to very. There seems to be variation on this matter,41 but k’aa by itself is not productively used for intensification. HH rejects k’aa in a translation of a sentence with very (89).

VG accepted (89), but another test for teasing apart superlatives from intensification shows that even for VG, k’aa seems to have superlative semantics. Following Hohaus (2015: 127-8), visual contexts were set up in which ‘very GA’ would be true of a given entity but ‘the most GA’ would be false due to the presence of another entity that instantiates a greater degree. VG rejects k’aa in such contexts regardless of whether the cardinality of the potential comparison class is two (90) or more (91).

Note that in both contexts, positive forms without k’aa are accepted. VG’s comment in (91) is also in line with a superlative analysis of k’aa. Understanding the full distribution of k’aa requires further investigation, and it is possible that multiple lexical entries will have to be postulated. I conclude from the available data that there must be at least an entry of k’aa with superlative semantics.

In Section 7, I demonstrated that k’aa can give rise to comparative and superlative readings even when the GA is minimum-standard. I suggested that k’aa should be analyzed as a superlative morpheme. My proposal is that in Gitksan, there is no LF distinction between comparatives and superlatives, that is, there is only one LF for a given construction involving k’aa, and it is compatible with both contexts in which comparatives and superlatives would be used in English. The only difference between what I have called comparative and superlative ‘readings’ is the size of the comparison class in the context.

Recall the analysis of GAs proposed in Section 5, where, due to the scale-sensitivity of the standard function, relative GAs (92 repeated from 66) have context-sensitive standards, while minimum-standard GAs (93 repeated from 44) denote context-dependent relations between degrees and individuals.

I analyze k’aa as a superlative morpheme (94), adapting Heim’s (1999) analysis of -est. It takes a GA denotation G and returns a function of the same type; the new function takes a degree d and an individual x and returns true iff G holds of d and x and for all non-x individuals in the comparison class g_c(C₃), G does not hold of d and y.

One difference from -est is that k’aa does not existentially close the degree argument d; this is because it is compatible with an MP (95).

To illustrate, (96) in both comparative (82) and superlative (87a) contexts receives the denotation in (97), again assuming the presence of an existential closure operator (53) in the absence of an MP.

The resulting interpretation is identical to that of superlatives in English, except that the comparison class g(C₃) can have a cardinality of two, in which case a comparative reading arises.

When the GA is relative, the existentially closed degree argument is the degree of deviation from the contextual standard. For example, (98), repeated from (84), receives the denotation in (99).

(99) is true iff there is a degree d such that Mary exceeds the standard degree of tallness by d, and no other member of the comparison class exceeds the standard degree by d. I assume that the GA and k’aa reference the same comparison class, that is, g_c(C1)(Mary), g_c(C1)(y) for any y∈g_c(C3), and g_c(C3) refer to the same set. Crucially, despite the degree d corresponding to deviation from the standard, k’aa constructions are not predicted to have an evaluative interpretation; (99) does not have an inference that Mary is considered tall in general. This is because, in contexts where the superlative morpheme k’aa is felicitous, there is always a specific comparison class provided by the context. In the context in (98), the comparison class contains only Mary and John, just like in positive constructions without k’aa (see Section 5). As a result, the output of the standard function is John’s height. What the sentence ends up asserting is that Mary exceeds John’s height by some degree but John himself does not, which is equivalent to Mary being taller than John.

Before concluding, I will argue that when relative GAs without the superlative morpheme k’aa are accepted in comparative contexts (Section 4), one factor that helps narrow down the value of the standard degree is a consideration of alternatives.

For example, in (100) (repeated from (73)), even without the standard phrase, the fact that the speaker did not choose to predicate ’wii ’nakw ‘tall’ of Lisa in the context in which Michael and Lisa are the only salient individuals gives rise to an inference that the denotation of ’wii ’nakw ‘tall’ ((101), repeated from (92)) must only hold of Michael and not Lisa. Therefore the standard degree must be a salient height degree that Michael exceeds but Lisa does not, namely Lisa’s height.

There are two pieces of evidence that alternatives play a role in comparatives in Gitksan. First, translations of both comparatives and superlatives from English optionally involve focus extraction. Gitksan (and the broader Tsimshianic family) has A’-extraction patterns shared by wh-questions, relative clauses, and focus marking (Rigsby 1986, Tarpent 1987, Davis & Brown 2011, Forbes 2012, 2018, Aonuki 2022, Brown 2023, 2024a, b). Extraction morphology varies by the syntactic position of an extracted element, and intransitive subject extraction is marked by a suffix -it, as illustrated by focus extraction in (102).

This A’-extraction aids comparative and superlative readings for VG (103). For example, while (103a) is rejected, extracting Mary in the second clause brings out the intended comparative reading (103b).

Similarly, while the positive form in (104a) was rejected with an intended superlative reading, VG repaired it by extracting the associate of comparison (104b).

That comparative and superlative readings aided by focus extraction in (103b) and (104b) are due to the context-sensitivity of relative GAs rather than, say, a covert comparative or superlative operator, is supported by the unavailability of such readings when focus extraction is applied to analogous sentences with minimum-standard GAs (105–106).

Second, translations of comparatives and superlatives from English are often volunteered with a morpheme gay ‘instead.’ Its uses outside of degree constructions (107) show that gay can associate with any lexical element in the sentence and signal that there is a salient alternative to the referent of the associate that makes the proposition false.43

Evidence that gay ‘instead’ is not a degree operator again comes from scale-structure differences. When the GA is relative (as in (108) and (109)), both forms with and without k’aa are volunteered in the presence of gay.44

Similar to extraction, gay ‘instead’ seems to help bring out the meaning of comparison. Adding gay to the rejected example in (103a) improves acceptability for HH (110).

On the other hand, when the GA is minimum-standard, the superlative morpheme k’aa is obligatory, and gay alone is not sufficient in either comparative (111) or superlative (112) contexts.

In this section, I have argued that what helps narrow down the value of the standard degree and gives rise to comparative and superlative readings without k’aa is a consideration of alternatives in the comparison class. Two empirical facts in support of this view are optional uses of focus extraction and an alternative-sensitive particle gay. In the absence of an overt superlative morpheme k’aa, both strategies are only available for relative GAs and not minimum-standard GAs. The incompatibility with minimum-standard GAs shows that the relative GA counterparts do not include a covert comparative operator akin to -er and that instead, the comparative and superlative readings are due to the context-sensitivity of relative GAs.

The observation that consideration of alternatives helps narrow down the context-sensitive standard of a relative GA and gives rise to comparative readings is not new. Bochnak (2013) reports that cleft constructions involving relative GAs in Luganda receive comparative interpretations (113).

In English, definite descriptions have been reported to be able to shift which entities satisfy a given relative GA (Kyburg & Morreau 2000). For example, Syrett et al. (2004) provide experimental results showing that, in a context with two entities, even if neither of them is independently considered [GA], a request like Please give me the [GA] one successfully solicits the [GA]-er of the two objects if the GA is relative, such as big and long. As Syrett et al. (2004) argue, shifting of the standard so that one and only one of the two objects satisfies a relative GA is driven by the existence and uniqueness presuppositions of the definite article. Gitksan thus adds to the cross-linguistic evidence that consideration of alternatives helps determine or shift the standard degree of relative GAs.

I have argued that GAs in Gitksan have inherently context-sensitive denotations, as illustrated by the denotation of ’wii ’nakw ‘tall’ in (114), repeated from (101).

A theoretical implication of this proposal for degree semantics cross-linguistically is that inherent context-sensitivity of GA denotations is not an exclusive property of languages without degrees. Following Bochnak et al. (2020), I will refer to languages that have been argued to lack degrees as a semantic primitive (Beck et al. 2009, Bochnak 2013, 2015, Bowler 2016, Deal & Hohaus 2019) as ‘degreeless,’ and those that have been analyzed with degrees as ‘degreeful.’45 Prior to Oda’s (2008) proposal for inherently context-sensitive GA denotations for Japanese (though see the issues raised for Japanese later in this section), the state of the literature was such that inherent context sensitivity was a feature of a degreeless language (Klein 1980, 1982, 1991). GA denotations in degreeful languages were necessarily context-independent, with context dependency contributed by additional (covert) morphemes or derived as an implicature (Bartsch 1975, Cresswell 1976, von Stechow 1984, Kennedy 1999, Rett 2007, 2008).

However, the typological picture resulting from this paper is that inherent context sensitivity is a point of cross-linguistic variation among degreeful languages (Table 1).

The prediction of the degreeful and inherently context-sensitive denotation in (114) is that positive constructions, that is, constructions without an overt comparative or superlative operator, should exhibit a mix of the properties that are traditionally associated with explicit and implicit comparatives (Kennedy 2007a). First, assuming that the context is rich enough to narrow down the comparison class g_c(C₁)(x) to a set consisting only of x and another individual that serves as the standard of comparison, the denotation in (114) predicts that there should be a comparative reading available without an overt comparative morpheme or an overt standard when the GA is relative. Such readings should be felicitous in crisp judgment contexts, given that the GA denotation does not require the degree argument to stand out from the contextual standard, unlike pos (Graff 2000, Kennedy 2007a, b, 2011). Second, unlike implicit comparatives in both degreeful and degreeless languages, comparatives formed with a GA like (114) should be compatible with an MP, and the resulting interpretation should be differential, given that the degree argument of (114) corresponds to the gap between the associate’s degree and the standard degree. Third, if a comparative reading available with a relative GA like (114) is indeed due to the GA’s context-sensitivity rather than a covert comparative operator, an analogous construction involving a minimum-standard GA should not be able to receive a comparative interpretation. In summary, inherently context-sensitive denotations of GAs like (114) predict the existence of a comparative strategy that is compatible with relative GAs even in crisp judgment contexts and can host a differential MP but is nevertheless incompatible with minimum-standard GAs.

This is exactly what we saw in Gitksan. Table 2 shows properties of the two constructions that can receive comparative interpretations in Gitksan, namely positive constructions and constructions with an explicit superlative operator k’aa (see Section 8), along with the properties of canonical implicit and explicit comparatives. While k’aa constructions align with canonical explicit comparatives, positive constructions show mixed properties of implicit and explicit comparatives.

How about Japanese? Japanese motivated the original proposal that a degreeful language can have inherently context-sensitive GA denotations (Oda 2008).47 We have already seen that relative GAs can directly combine with MPs without an overt standard, and the resulting MP interpretation is differential (see (115), repeated from (77)).

Can a relative GA receive a comparative interpretation without an MP? The answer likely varies by dialect, but it is not as widely or easily available as the inherently context-sensitive approach to GAs predicts. At least in my own judgments in the Tokyo dialect, (116) does not have the intended comparative reading and rather sounds contradictory to the preceding sentence.48

In contrast, recall that an analogous example in Gitksan was even volunteered as a translation of a comparative (the second clause of (117), repeated from (55)).

Adding an alternative-sensitive particle hoo to (116) can bring out a comparative reading (118).49

Given the contrast between (116) and (118) alone, it is conceivable that the GA takai ‘tall’ is inherently context-sensitive, and the comparative reading in (118) arises from the reasoning triggered by the alternative-sensitivity of hoo that takai is not true of the other entity, the maple tree (see Section 9).

However, crucially, a parallel construction with a minimum-standard predicate can also receive a comparative interpretation (119a), although it is ambiguous with the reading in which only one entity satisfies the GA (119b).50

To the extent that (119a) is acceptable with a comparative reading, the possibility that there is a covert comparative morpheme akin to -er in (118) cannot be completely ruled out.

There are comparative constructions that are incompatible with minimum-standard GAs in Japanese (Sawada 2009, Pearson 2013), which involve a conditional antecedent ‘if compared to x.’

However, as Sawada (2009) demonstrates, they pattern with canonical implicit comparatives (Kennedy 2007a): they are also incompatible with MPs and crisp judgment contexts (see Section 2.3). The existence of canonical implicit comparatives cannot be predicted by the existing proposals for inherently context-sensitive denotations of GAs in Japanese (Oda 2008, Aonuki 2024c) nor by applying the current proposal for Gitksan to Japanese, at least without additional mechanisms.51

Finally, the approach of assigning inherently context-sensitive denotations to GAs has been applied to English as well, in an attempt to account for distribution of evaluativity inferences (Breakstone 2012) and as an alternative to postulating pos (Cariani et al. 2023, 2024, Wellwood 2014). The potential for over-generating comparative readings in the English counterparts of positive constructions like (116) seems to be a shared issue that should be addressed in these analyses as well.52

This paper demonstrated that, in Gitksan, Measure Phrases (MPs) can co-occur with relative gradable adjectives (GAs) without a comparative morpheme and consistently receive differential interpretations, while minimum-standard GAs give rise to absolute interpretations of MPs. The former has only been observed in Japanese and Korean, to my knowledge.53 Unlike in Japanese, the distinction between relative and minimum-standard GAs obtains even in the absence of MPs, that is, relative GAs can receive comparative and superlative interpretations without an overt comparative morpheme, while minimum-standard GAs cannot. I argued that these two scale-sensitive phenomena receive a unified explanation in an approach that assigns inherently context-sensitive denotations to GAs, initially developed for Japanese by Oda (2008) and later adopted for English by a minority of analyses (Breakstone 2012, Cariani et al. 2023, 2024, Wellwood 2024). I proposed inherently context-sensitive denotations of GAs in Gitksan, which builds on Oda (2008) and makes explicit predictions about when comparative readings as opposed to positive readings arise.

I also revisited Japanese and observed that unlike in Gitksan, the possibility of an alternative analysis, namely that there is a covert -er in Japanese, cannot be completely ruled out. Gitksan shows what a language with inherently context-sensitive GAs would look like empirically. To the extent that the current analysis of Gitksan is successful, it raises a question of why other languages that have received similar analyses exhibit only a subset of the empirical properties predicted by inherent context-sensitivity.

The latter half of the paper investigated the semantics of an overt degree morpheme k’aa and argued for its analysis as a superlative morpheme. I also argued that in achieving comparative and superlative readings without k’aa, consideration of alternatives to the subject plays a role, as also observed in other languages (Kyburg & Morreau 2000, Syrett et al. 2004, Bochnak 2013).

Finally, the MP interpretation data in Gitksan are also significant from the typological perspective. The existence of typologically unrelated languages showing the same pattern suggests the possibility that the pattern is not actually as rare as previously assumed. Nevertheless, there are claims that GAs being able to combine with MPs without a degree operator is an exception rather than the norm (Schwarzschild 2005). Is the pattern of differential MP interpretations shared by Gitksan and Japanese indeed typologically rare, and if so, why? The former question can only be addressed by further cross-linguistic investigations.