The Development of Polarity Subjunctive
📚 Introduction
- In Modern Spanish the verb saber `know’ usually selects indicative:
- However, when negated, the embedded verb can take indicative or subjunctive mood, a phenomenon known as Polarity Subjunctive (Quer 1998):
📚 Presuppositional Account
- Indicative presupposes that the speaker is committed to p, whereas subjunctive does not trigger such presupposition (Alonso 1999):
| Embedded Mood | Asserts | Presupposes |
|---|---|---|
| ind | Ariel didn’t know it was warm | \(\leadsto\) it was warm |
| subj | Ariel didn’t know it was warm | – |
📚 The First Person Effect
- No Mood Alterntion1: 1st person matrix subject + present tense matrix verb (Alonso 1999, 3222–23; Quer 1998, 61–62)
Presupositional Account: indicative in (3) is ungrammatical because it leads to a semantic contradiction
- Sentence (3) asserts: I don’t believe it is raining
- Sentence (3) with indicative presupposes: I believe it is raining
⚠️ A Problem with the Presuppositional Account
- Recent experimental data (Montero and Romero 2023) shows that the presuppositional difference only emerges (is significant and large size) with cognitive-factive verbs:
V1 (cognitive-factives): know, notice, remember, see and find out.
V2 (non-factive/non-fiction): believe, think, say, tell and suspect.
V3 (fiction): dream, fantasize, invent, fake and make believe.
💡 Towards a new proposal
What if the alternation does not have a clear semantic contrast? How can we prove that?
Evidence of absence fallacy: experimentally one can prove the existence of a given meaning, but cannot prove the absence of meaning.
Variational Specialization: (Wallenberg 2019)
📉 Corpus Study
Search 
- not [believe/say/know/see] that
- 12th-20th
- published in Spain (to avoid dialectal effects)
- Total tokens: 12.360
Annotation
- Mood in the embedded clause
- Matrix Subject (first or other)
- Tense of the matrix verb
- Mood matrix verb
- declarative/interrogative
📉 Results
⚠️ Potential Problem
- For the verb believe although indicative is dropping after 1600, there is an increase in the use of indicative during 1200-1500\(\rightarrow\) Looks more like a failed change !
Failed Changes
- Postma (2010) and Bacovcin (2017) have proposed to analyze failed changes as the result of two opposing successful changes (multiplication of logistics):
- Contact with Andalusi Romance/Arabic: goes from the conquest in 711 to the expulsion of the Moriscos in 1610 (Ruhstaller and Gordón Peral 2017).
How can we formally charaterize the alternation after non-factive verbs?
💡 Competing Grammars (Kroch 1989)
- G1: Semantics accounts like Schlenker(2003) and Giorgi \(\&\) Pianesi(1997) \(\longrightarrow\) indicative
- G2: Subj is the equivalent of genitive/partitive case at the clausal level (Artiagoitia and Elordieta 2016; Uriagereka 1988; Kagan 2012)
📚 G2: Mood as a Case Marker
- Genitive of Negation in Polish (Błaszczak 2001)
- Parallelim between GoN and Mood (Kagan 2012):
📚 Genitive of Negation
- This phenomenon has been analysed as a Case-Agreement relation (Błaszczak 2001). If polarity is positive the DP is valued as accusative, if negative as genitive:
📚 G2: Mood as a Case Marker
- Since clauses also trigger verb-subject agreement, we argue they also carry i\(\phi\)-features:
- [[That the march should go ahead] and [that it should be canceled]] have been argued by the same people at different times. (McCloskey, 1991:564)
- The derivation would proceed as follows:
What causes the First Person Effect with the verb believe?
💡 Multiple linguistic changes
- Linguistic Change 1: type of mood selection
- Linguistic Change 2: where the CP is merged in the structure (Torrego and Uriagereka 1992)
💡 Multiple Linguistic Changes
Change in Complementizer-drop only affected first person constructions
💡 Multiple Linguistic Changes
Change in Complementizer-drop only affected 1st person constructions
🖥️ Rate of Change: First Approximation
(Postma 2010) and (Bacovcin 2017): failed changes can be modeled as the combination of two logistics
Simplified version: failed change is the combination of two exponentials (Laplace)
\[M(t)=\frac{1}{1+e^{-\frac{t-to}{s}}} \cdot \frac{1}{1+e^{-\frac{t-to'}{s'}}}\]
\[M(t)=q\cdot e^{-\frac{|t-to|}{s}}\]
🖥️ Model Fitting (nls)
First Person: \(1\cdot e^{-\frac{|x-1543\pm8|}{143\pm14}}\)
Non-first Person: \(0.695\cdot e^{-\frac{|x-1549\pm40|}{373\pm134}}\)
- First Person: \(0.430\cdot e^{-\frac{|x-1578\pm18|)}{194\pm44}}\)
🖥️ Adding the Effects:
\(\underbrace{q\cdot e^{-(\color{blue}{\overbrace{\frac{|t-to_c|}{s_c}}^{CP-merge}}+\color{green}{\overbrace{\frac{|t-to_b|}{s_b}}^{selection-type}})}}_{\color{red}{derived-model}}\) \(\simeq \underbrace{q\cdot e^{\frac{|t-to_a|}{s_a}}}_{original -model}\)
Making the assumption that \(|t − to_c| ≈ |t - to_b|\), ( these values differ 30 years which given our scale can be considered a small difference), we obtain:
where \(to_{bc}\) is the mean between \(to_b\) and \(to_C\). By the general rules of rational functions we obtain:
Given that we have multiplied by n and our slope has the form 1/n, we need to dive by it again, leaving us with the following equation:
Then we just have to input our values: \(s_c\)=194, \(s_b\)=373, \(to_{bc}=1563\)
As can be seen the obtained slope is very similar to the one obtained from fitting the original data: 128 \(\simeq 143 \pm 14\)
The value of q for plotting is the initial condition we are trying to model (in our case will be the one for First Person constructions which was equal to 1).
Thank you for your attention!
Thanks to the team: George Walkden, Henri Kauhanen, Gemma McCarley, Molly Rolf and Sarah Einhaus. https://www.ling.uni-konstanz.de/en/walkden/starfish/
We gratefully acknowledge funding from the European Research Council, grant no. 851423 (STARFISH)
