2.1 Two possessive paradigms

Old French features two morphologically distinct paradigms of prenominal possessive morphemes. The two paradigms are illustrated for the first person possessives in Tables 1, 2 from Buridant (2019: 219). We label the paradigms “short” and “long” as pre-theoretical descriptions referring only to their relative phonological weight.³ By the end of the fourteenth century, long forms virtually go out of use.⁴

We investigate the distribution of the two series in the treebanks of Martineau et al. (2010) and Kroch & Santorini (2021).⁵ We extracted from these corpora all noun phrases with possessives in 1st, 2nd, or 3rd person singular (N = 25,104 noun phrases), since only these distinguish between short and long forms. Figure 1 shows the relative frequency of the two paradigms across time.

Figure 1

Estimated probability of short vs. long possessives in French.

There is no consensus in the literature as to whether or to which extent the morphological contrast is a reflection of syntactic and semantic differences. Gamillscheg (1957) argues that the choice between the paradigms is governed by metrical considerations only.⁶ Arteaga (1995) notes that the two types can be coordinated, which suggests their syntactic and semantic equivalence. In contrast, Butet (2018) and Buridant (2019) argue that the two series differ with respect to the possibility of co-occurrence with determiners: while short forms tend to not co-occur, long forms tend to do so. Butet (2018) takes the co-occurrence to be a hallmark of an adjectival status, and absence of co-occurrence as a signature of the determiner status. Our corpus data show that while both types of possessives do occur with and without determiners, diachronic trajectories differ dramatically.

2.2 Co-occurrence with determiners

Examples (7) and (8) illustrate co-occurrence of a short form tos and a long form tuen, respectively, with an l-determiner (i.e., le/la/les).⁷

Examples (9) and (10) illustrate co-occurrence of short and long forms, respectively, with indefinite determiners.

In Modern French, co-occurrence of (short) possessives with any determiner type is strictly ungrammatical, as (11) and (12) illustrate.

Modulo the claim by Butet (2018) that short and long possessives co-occur with determiners at different frequencies, it has been commonly assumed that the general trend in the co-occurrence pattern is that of a continuous decrease in the frequency of determiners in noun phrases with prenominal possessives. As mentioned above, this has been taken to indicate a progressive reanalysis of prenominal possessives from adjectives, capable of co-occurrence, to determiner heads, incompatible with another determiner. There has been however no large-scale quantitative studies testing the original empirical assumption underlying the reanalysis view. In order to carry out such a quantitative study, we fit logistic regression models of the form in (13) to three datasets: noun phrases with short possessives (N = 24,607 noun phrases), noun phrases with long possessives (N = 497 noun phrases) and noun phrases without possessives (N = 186,768 noun phrases).

The logistic function, which describes processes that have phases of a slow take-off, rapid growth, and again a slow attenuation, has been identified as a model particularly suitable for capturing successful language changes (see Altmann et al. (1983); Kroch (1989); Niyogi & Berwick (1997); Kauhanen & Walkden (2018) for details). In our case, the model predicts the probability that the binary variable Determiner takes on the value yes given variable Date as a predictor. The coefficient β reflects the importance of the time factor for predicting the determiner probability, while α corresponds to the predicted probability at an (idealized) time point 0. By estimating the parameters of a model fitted to our data on the determiner-possessive co-occurrence and also by comparing the performance of this predictive model and an alternative one which describes non-monotonic processes, we can evaluate to which extent the assumption about a steady co-occurrence decrease is reliable.

The three logistic regression models are plotted in Figure 2 together with data points which correspond to the empirical frequencies of determiners in the three types of noun phrases in 50 year intervals.⁸ For model fitting, we used a normalized time period variable, to facilitate comparisons across languages, since different time spans are available for observation in each case.

Figure 2

Determiners in noun phrases with and without possessives in French, logistic models.

The parameter estimates for the three models are given in Table 3.

The model fit to the dataset without possessives detects a significant rise in the use of determiners (time coefficient = 0.1221, p < 0.001***). This means that the likelihood that the perceived diachronic trend is due to chance and that the weight of the time factor is zero is very small. That the coefficient is positive means that that for the higher values of date the model predicts higher probabilities of determiner appearance or, put differently, that the frequency of determiners increases over time in noun phrases without possessives. This matches the results presented in Simonenko & Carlier (2020b) on the general increase in the frequency of determiners in historical French.

In the dataset with long possessives we find a similar trend. The frequency of determiners in noun phrases with long possessives increases in Old French at a statistically significant positive rate (time coefficient = 0.9064, p < 0.001***). This means that at least for the long possessives, the assumption about co-occurrence decrease is wrong. It appears that the frequency of determiners in noun phrases with long possessives goes towards its maximum about the same time that they disappear from the language.

The dynamics of the co-occurrence pattern in noun phrases with short possessives are more complex. In this case, the model detects a significant decline in determiner use (time coefficient = –1.2242, p < 0.001***). On the face of it, this result matches the assumption about a steady co-occurrence decrease. In the following section, however, we present both statistical and linguistic arguments suggesting that a monotonic decrease assumption may be erroneous for the short possessives as well.

2.3 A failed change

Although the model fitted to the noun phrases with short possessives detects a diachronic trend, it does not fit the data particularly well for the texts composed before 1000. Moreover, the distribution of data points that correspond to individual texts (in black in Figure 3) makes one suspect a different, non-monotonic trend, namely, the rise of the co-occurrence frequency before 1000 and a decline afterwards. Such a trend is entirely expected given the general philological reasoning. While Classical Latin did not have a consistent marking of (in)definiteness, from Late Latin onwards, definiteness marking at the noun phrase level increases (Carlier & De Mulder 2010). This increase continues in Old French, in the form of the definite determiner derived from Late Latin demonstrative ille (Simonenko & Carlier 2020a). It is therefore highly plausible that, in comparison with the Classical and even Late Latin baseline, the stage of the determiner loss in noun phrases with short possessives that we clearly observe in Old French after 1000 has been preceded by their rise in this context before 1000. We therefore suggest that instead of a steady decline in the frequency of short possessive-determiner co-occurrence, a rise-and-fall or a bell-shaped pattern should be assumed. The rise-and-fall evolution hypothesis is all the more plausible in view of the fact that the possessive only rarely co-occurs with a (demonstrative) determiner in Latin (including in Late Latin). In general, the rate of bare nouns in Classical Latin has been estimated to be 77% (Carlier & Lamiroy 2018), while at the beginning of the Old French period it is down to approximately 40% (Simonenko & Carlier 2020a). The empirical question at stake here is whether the rise-and-fall pattern can be detected in the observable time period, that is, after 800, or whether the rise of determiners with short possessives precedes the earliest preserved records in the vernacular. In what follows, we will formally evaluate the assumption that the rise-and-fall can be observed in the available data.

An initial rise in the frequency of a particular grammatical phenomenon followed by its decline has been referred to as a “failed change” pattern since Postma (2010). A failed change is not a particularly rare phenomenon. On the basis of data presented in Oliveira e Silva (1982), Postma (2010) argues that in Brazilian Portuguese, the frequency of determiners in noun phrases with possessives first rises and then goes down between 1650 and 1850. Another example is the failed do-support in positive declarative contexts in English. This construction grows in frequency from around 1500 to 1560 and rapidly declines afterwards (Postma 2010). The distribution of to-marking with the recipient argument in English ditransitive constructions has likewise been analyzed as a failed change by Bacovcin (2017). Truswell et al. (2019) describe the pattern characterizing the use of the ne … not construction from Early Middle English until the fifteenth century as a failed change.

Postma (2010) proposes to use a bell-shaped or parabolic function of the type in (14) to model such failed changes.⁹

If a parabolic function in (14) fits the data on the determiner (non)occurrence in the set of noun phrases with short possessives as good or better than a simple monotonically decreasing logistic function, we can conclude that the failed change assumption for the evolution of determiners with short possessives in the observable data window is as viable or more viable than the simple decline assumption.

To evaluate a parabolic model, we fit it to the dataset consisting of noun phrases with prenominal short possessive forms. The result is illustrated in Figure 3, where green points correspond to 50 year bins (used for model fitting) and black ones – to determiner frequencies in individual texts, which are unevenly distributed across centuries.

Figure 3

Determiners in noun phrases with and without possessives in French, parabolic model added.

The parameter estimates for the parabolic model are given in Table 4.

Sparsity of determiner occurrences in noun phrases with short possessives in general and before 1000 in particular makes it difficult to compare the simple logistic vs. parabolic models. The two earliest texts, which do not contain any determiners in noun phrases with short possessives, are very short. The first one, Sermons de Strasbourg (composition date 842 A.D.), consists of 115 words and contains 2 noun phrases with short possessives, as in (15), and 4 with long, of which 1 with a determiner, (16)–(17) (second noun phrase) and 3 without, (17) (first noun phrase).

The second earliest text, Séquence de sainte Eulalie (composition date 881), counts 189 words and has only 1 noun phrase with a short possessive and without other determiners, (18). It also has 2 noun phrases with long possessives, one with, (19), and one without a determiner, (20).

Although absolute numbers of noun phrases with either short or long possessives in these earliest texts are very low, we nevertheless do see determiners with the long, but not with the short paradigms.

We now proceed to the evaluation of both models in terms of classificatory accuracy. In particular, we compare the predictive performance of simple logistic and parabolic models using their receiver operating characteristics (ROC), which correspond to the relation between model’s sensitivity (proportion of true positive cases predicted to be positive among all true positive cases) and specificity (proportion of true negative cases predicted to be negative among all true negative cases) across multiple possible cut-off points, that is, different arbitrary thresholds for predicted probabilities below which a case is classified as negative and above which it is classified as positive. To compare models’ ROCs, we use the area under the curve measure (AUC). A greater AUC indicates a greater predictive accuracy of a model across multiple cut-offs. ROC (AUC) is 0.7115 for the logistic model corresponding to the steady decrease assumption within the observable window, and 0.7116 for the parabolic model, which assumes a rise-and-fall pattern within the attested data window. The models’ performance proves to be essentially identical and in both cases quite good. The search for a better model will not be pursued further here.¹⁰ However, this result allows us to conclude that a parabolic or rise-and-fall model is at least as good at capturing the distribution of determiners with short possessives as a steady decline model. We would like to stress that the rise-and-fall evolution hypothesis is the only one which is able to reconstruct the gap between Late Latin and Old French, and the question we have just explored is whether this pattern manifests itself in the observed data window. We conclude that it is plausible that it does, namely, that the available Old French sources offer evidence of a failed change in the context of short possessive noun phrases.

In Postma’s perspective, a change can be “doomed” to fail from the outset: the frequency of certain forms rises only to go down in the next generation because of an inherent suboptimality of the configuration resulting from the change. On this view, which we adopt, short possessive forms intrinsically have a property (or properties) making their co-occurrence with overt determiners suboptimal. We argue that this property is their determiner-like semantics, which enables them to spell out D (details are given in section 5).

3.1 Two possessive paradigms

Medieval Spanish features two paradigms of possessives, short and long forms, both of which can occur in the prenominal position. Sub-paradigms of both types for the first person are given in Tables 5, 6.

Similarly to French (cf. Figure 1), long forms disappear from the prenominal position, as illustrated in Figure 4. Our dataset consists of 1st, 2nd, and 3rd person short and long prenominal possessives (N = 1,275,626 noun phrases, of which 3657 have long possessives) taken from Davies (2006-).

Figure 4

Estimated probability of short vs. long possessives in Spanish.

3.2 Co-occurrence with determiners

In medieval Spanish, both short, (21)–(22), and long, (23)–(24), series can be used with and without determiners in the prenominal position (Labrousse 2018: 38).

In Modern Spanish, short forms cannot co-occur with determiners, while long forms, restricted to the postnominal position, can, as illustrated in (25)–(26).¹¹

Below it is shown that, just like in medieval French, the frequency of determiners increases in medieval Spanish with long forms (until the long forms disappear from the prenominal position), as well as in noun phrases without prenominal possessives, and decreases with short forms. Again, as in medieval French, we show that it is very plausible that the frequency of determiners in the latter case follows an “up-and-down” pattern and thus constitutes a case of failed change.

In order to evaluate the plausibility of the monotonic decrease vs. failed change assumptions for the short possessives, we first fit logistic regression models of the form in (13) to three data sets: noun phrases with (mostly) short possessives (N = 7703 noun phrases from Labrousse 2018: 79), noun phrases with prenominal long forms from Davies (2006-) (N = 3657 noun phrases), and noun phrases without possessives from P.S. Post Scriptum (2014) corpus (N = 6121 noun phrases).¹² These models are graphed in Figure 5, and the parameter estimates are given in Table 7.

Figure 5

Determiners in noun phrases with and without possessives in Spanish, logistic models.

The models detect a negative trend with short forms (time coefficient = –0.0084, p < 0.001***), a positive trend with long ones (time coefficient = 0.001, p < 0.001***), and a positive trend in noun phrases without possessives (time coefficient = 0.0014, p < 0.001***).¹³

3.3 A failed change

We further evaluate the assumption that in Spanish noun phrases with short possessives, an up-and-down pattern is observable in the available data window. As in the case of French, the hypothesis of an initial rise of determiner frequency, before its decline, is the only one which is able to reconstruct the gap between Late Latin and Old Spanish. To see how plausible this assumption is, we fit a bell-shaped model of the type in (14) to this dataset. The latter replaces the logistic model in Figure 6. Data points for individual texts (in black) have been added here, in addition to 50 year bins (in green). Parameter estimates of the parabolic model are given in Table 8.

We compare the predictive accuracy of the two models, s- and bell-shaped, using ROC (AUC) characteristics. For the s-shaped or logistic model ROC (AUC) equals 0.70 and for the bell-shaped model – 0.76, which indicates a somewhat better performance of the latter. Hence, the assumption about the evolution of determiners with short possessives following an up-and-down trajectory within the observable data window is no less (or slightly more) plausible than the monotonic decrease assumption.

Figure 6

Determiners in noun phrases with and without possessives in Spanish, parabolic model added.

4.1 Two possessive paradigms

Short and long prenominal possessive paradigms from medieval Portuguese are illustrated in Tables 9, 10 for the first person.¹⁴

In contrast to French (and Spanish prenominal position), it is the short forms that do not survive. We track the distribution of the two types in the prenominal position (N = 324,599 noun phrases) using the corpus of Davies (2002-). The results are visualized in Figure 7.

Figure 7

Estimated probability of short vs. long possessives in Portuguese.

4.2 Co-occurrence with determiners

In historical texts, both long and short forms are attested with and without determiners in the prenominal position. Examples (27)–(28) and (29)–(30) offer illustrations for long and short forms, respectively.

In Modern European Portuguese, in the prenominal position, long (and only long, short forms having disappeared) possessives normally co-occur with definite determiner, whereas in the postnominal position they are used with indefinite determiners, numerals, wh-words, as well as without determiners (cf. Brito (2007: 31), among many others).¹⁵

In order to evaluate the overall trends (increase or decrease) in the frequency of determiners and, in particular, the monotonicity assumption in the case of short possessives, we once again fit logistic regression models of the type in (13) to the datasets consisting of noun phrases with long prenominal possessives (N = 318,550 noun phrases, from Davies (2002-)), with short prenominal possessives (N = 6,409, from Davies (2002-)), and without possessives (N = 27,705 noun phrases, from Galves et al. (2017) and P.S. Post Scriptum (2014)). We visualize this in Figure 8.

Figure 8

Determiners in noun phrases with and without possessives in Portuguese.

As can be seen from Table 11, a logistic model detects a positive trend both in all three contexts, time coefficient being 0.5965 (p < 0.001***) with long possessives, 0.1881 (p < 0.001***) in noun phrases without prenominal possessives, and 0.1107 with short possessives (p = 0.0037**). In the latter case, the model’s estimates are the least reliable, especially after 1600. As Figure 7 indicates, around this time short possessives become vanishingly rare in the language.

The use of determiners with short possessives, however, seems to substantially differ from their distribution in noun phases with long possessives. We fit a bell-shaped model in (14) to the determiner distribution with short possessives in order to evaluate the plausibility of an up-and-down change trajectory in the context of short possessives. The result is plotted in Figure 9 and the parameter estimates of the parabolic model are given in Table 12.

Figure 9

Determiners in noun phrases with and without possessives in Portuguese, parabolic model added.

Comparing the two models, logistic and parabolic, in terms of their predictive ability as measured by ROC (AUC), is not very informative in this case: both perform very poorly, at around the chance level, with ROC (AUC) being 0.52 for the former and 0.41 for the latter. While evaluating the two models for the context of short possessives in Portuguese, we need to keep in mind that unlike the data on short possessives in French and Spanish, this dataset comes from a corpus without syntactic annotation and contains a degree of noise. The exploration of the determiner use trends in noun phrases with short possessives in Portuguese remains somewhat inconclusive, except for the important quantitative fact that their distribution in this environment clearly differs from the trend observed in the context of long possessives.

4.3 A comparison between French, Spanish, and Portuguese

For the ease of comparison of the evolution of determiner distribution in possessive and non-possessive noun phrases in the three Romance languages under consideration, we present the corresponding plots side by side in Figure 10.

Figure 10

Determiners in French, Spanish, and Portuguese.

These Figures for the three Romance languages under study visualise several important and previously unreported empirical results.

First, the determiner frequency grows during at least a certain period in all environments we have considered, that is, with short and long possessives, as well as in noun phases without possessives. In case of long possessives and noun phrases without possessives, the monotonic growth period correspond to the whole observed period. For the short possessives, we showed that it is likely that there is an initial growth period followed by the period of decline (except for the inconclusive results on Portuguese, where neither monotonic nor a bell-shaped models perform very well).

Second, the development in the context of long possessives seems to be largely in line, in terms of the general temporal trends rather than absolute rates, with the evolution of determiners frequencies in noun phrases without possessives. Moreover, in Spanish and Portuguese the two seem to follow a roughly parallel course (red and blue lines in the figures above).¹⁶

The third important result is that there is a strong difference in the trends observed in the environment of long prenominal possessives vs. short ones. While in the former environment the frequency of determiners clearly grows with time, in the latter it either declines or follows the up-and-down pattern (French and Spanish). Again, the quality of the Portuguese data on short possessives prevents us from deciding between a logistic model predicting a very slow growth or a model corresponding to a bell-shaped pattern. In either case, in Portuguese as well, the difference between the two environments is indisputable.

The fourth relevant observation that emerges from language comparison is that changes in determiner frequency in the context of long forms precede changes in the context of short possessives.

The analysis that we develop in the following section builds on these new empirical findings, as well as on the general theoretical consideration we mentioned before, namely, that it is necessary to admit that the frequency of determiners in the context of short possessives must have increased before it began decreasing in order to account for the transition between Late Latin and early Romance languages, whether we can capture this quantitative trend on the basis of the available empirical data or not.

5.1 Grammatical shift

In order to account for the fact that across the three examined languages, the growth of determiner frequency is observed both in noun phrases without prenominal possessives and with long possessives, as well as, initially, with short possessives (as established on theoretical and partially empirical grounds), rather than postulating changes at the level of individual possessive forms, we propose, following Simonenko & Carlier (2020a), that there is a general rise in frequency of the grammar where (in)definiteness is obligatorily marked at the noun phrase level. In other words, a single shift happens: a replacement of a grammar parsing nominals with their modifiers as an NP structure, as in (33a), by a grammar which parses them as a DP, as in (33b).¹⁷

Our analysis of this shift is couched in the perspective of grammar competition which assumes that in a given population of speakers more than one syntactic or semantic analysis of an utterance can be available (e.g., Kroch (1989) as one of the foundational works and Pintzuk (2003) for an overview). If with time the probability of using one of the grammars grows and that of using the “competing” grammar correspondingly declines, we can speak of a language change. The grammar competition approach has already been engaged for explaining the diachronic rise in frequencies in the determiner use in historical Spanish (Batllori & Roca 2000) and French (Simonenko & Carlier 2020a). We assume that the increasingly frequent parsing of nominals as structures with a determiner layer is a consequence of an emerging pressure to morphologically mark existential presupposition or its absence at the noun phrase level because marking it at the sentential level by means of constituent order and/or prosodic means becomes progressively unavailable, as suggested by Simonenko & Carlier (2020b) for French, among others. We adopt the latter point about causality as a conjecture without trying to develop and prove it further here.

In the following section, we present our account of the semantic and syntactic composition of long and short possessives that is compatible with the quantitative empirical results we have obtained.

5.2 Semantics of long and short forms

On the basis of our quantitative results, we endorse the widely advocated view that short and long forms have different semantico-syntactic properties, without subscribing to the claim that either paradigm underwent a semantic evolution. We propose, in particular, that they differ minimally in the presence of the DP layer in their internal structures. This difference, we argue, holds across all time periods and is independent of the NP-to-DP grammar shift operating in the diachrony of the three languages. Nominal expressions with long and short forms correspond to the structures in (34) and (35), respectively, where the parts in bold correspond to the internal structure of a possessive morpheme.

Here R is a relational operator that introduces a possessive pronoun with index i in its specifier (cf. R in the semantics of demonstratives in Elbourne (2008) and in the semantics of possessives in Simonenko (to appear)). The semantics of R is given in (36), where P argument at the Logical Form level corresponds to the denotation of the nominal predicate and y to the denotation of the possessive index (a possessor individual). We also assume that R introduces a presupposition that there exist individuals with the nominal property P in the relevant situation of evaluation (for the discussion see Simonenko (2014)). R relates individuals with the property denoted by the nominal to a contextually given possessor by a possessive relation, broadly conceived.¹⁸

For D, we assume a Fregean semantics from Heim (2011) in its situational implementation, as in Schwarz (2009), whereby D denotes a function which takes a situation s, a property P, and returns a Sharvy’s maximal individual that has P in s, provided that there exists such individual in the relevant situation, as in (37).

There is non-quantitative evidence from Modern Portuguese, which retained long possessives in the prenominal position, that prenominal long possessives spell out a functional head rather than an adjectival phrase. Brito (2007) shows that in European Portuguese, prenominal possessives cannot be modified by “exclusion” adverbs (só, apenas “only, just”), (38), in contrast to postnominal possessives, (39).

Castro & Costa (2002) show that prenominal possessives in Portuguese cannot be coordinated. Coupling this fact with the assumption that heads cannot be coordinated (Kayne 1994), they propose these forms spell out heads, rather than phrases.²⁰

Our proposal also captures the fact that in Modern Portuguese, in the prenominal position long possessives occur virtually exclusively with definite determiners. Since R introduces a presupposition that the extension of RP is not empty in the relevant situation, RP, if felicitous in a given context, satisfies the existential part of the presuppositional requirements of a definite determiner (the other part being the maximality requirement), as given in (37).²¹ On the basis of the empirical data extracted from the corpus of Davies (2002-), we argue that this situation extends onto the Medieval period as well: the use of prenominal possessives with indefinites (relative to the overall use with indefinite, definite, and demonstratives determiners) was already very marginal in the twelfth century, 0.013, and never became higher. Given the Maximize Presupposition! reasoning, our proposal explains why prenominal long possessives occur almost only with a definite D, rather than an indefinite D.

Given the representation for nominals with short possessives in (35), it can be expected that determiners are never used in this context, since there is already a D head present as part of the internal structure of short possessives. Empirically, however, we observe the co-occurrence pattern for some period of time, which, we argued, likely followed a bell-shaped trajectory. Assuming that a lexical item can spell out a subset of the features it is specified with, as in the nano-syntactic approach to lexicalization (see Starke (2018) for a recent exposition), we propose that the short possessives are compatible with either of the spellout options represented in (40). On the first option schematized in (40a), they spell out both the RP and the DP layers, whereas on the second only the RP layer.²²

We propose that the second spellout option, in (40b), is not competitive, in the long run, in comparison to alternatives that are more economical in terms of processing: short possessives without overt determiners or long possessives with overt determiners. The reason is that in order to generate a sequence ‘determiner + short possessives + noun’, some type of lookahead or post-insertion syntactic realignment is needed to make a short possessive morpheme match only R instead of D-R when D is lexicalized by another determiner. Alternative spellouts, which is to lexicalize both D and R by a single item (i.e., ‘short possessive + noun’) without adding a separate determiner (and thus avoid readjustment) or to lexicalize R by a morpheme which can lexicalize only R and therefore, again, does not require readjustment (‘determiner + long possessive + noun’), are simpler in terms of processing and therefore, we argue, preferred options. Nevertheless, the force of analogy with the constructions ‘determiner + noun’ and ‘determiner + long possessive + noun’ drives the use of ‘determiner + short possessive + noun’ for some time, before it starts decreasing again.

The analogy-based argument, in addition to conceptual considerations, rests on the aforementioned quantitative fact that the rise in determiner frequency begins earlier in the context of noun phrases with long possessives than with short ones in the languages we investigated, as Figure 10 shows. From a diachronic viewpoint, the tension between analogy and economy is eventually resolved in favour of the latter in all three languages, and the frequency of ‘determiner + short possessives + noun’ goes down.