How we distinguish fact and fiction

From Science News, August 10, 2017: “More than 300 years ago, the philosopher René Descartes asked a disturbing question: If our senses can’t always be trusted, how can we separate illusion from reality? We’re able to do so, a new study suggests, because our brain keeps tabs on reality by constantly questioning its own past expectations and beliefs. Hallucinations occur when this internal fact-checking fails, a finding that could point toward better treatments for schizophrenia and other psychiatric disorders.”

How do natural languages distinguish between fact and fiction? They don’t really. But they provide many tools for reality checking. I probed into this topic in my 2015 contribution to the Rome Science Festival. Here is the abstract for the talk: “Languages do not care all that much about the difference between fact and fiction. Stories tend to be told as if they were reports of known fact. You may have read somewhere that Mr. Palomar was standing on the shore, looking at a wave. Was it in a newspaper? Well, no, it’s from a story by Calvino. [Il signor Palomar è in piedi sulla riva e guarda un’onda.] Calvino’s story doesn’t tell you that it is a piece of fiction. There are no grammatical categories to indicate the merely fictional. There are no fictional declensions or conjugations in natural languages. Every sentence in every language describes a wide range of possibilities. When Calvino writes that Mr. Palomar was standing on the shore, he evokes a set of possibilities. When Wikipedia tells me that Calvino was born in Cuba, it, too, evokes a set of possibilities. But it does more. It also implies that the world we live in is one of them. When it comes to how to approach reality, languages do care. They have countless ways to modulate, fine-tune, and calibrate claims about reality. In a quarter of all the world’s languages, you can’t say that a man was standing on the shore without mentioning your evidence. Was it just hearsay? Did you see it with your own eyes? Did you infer the man’s presence from the footprints in the sand? In those languages, grammar forces speakers to be upfront about their evidence, just as Italian or English forces us to use tenses to locate events in time. Once the evidence is on the table, we may want to say something about the strength of the conclusions we can draw from it. Might there be an iguana in the reptile house? Is it certain? Or is there only a slim chance? Is there a better chance for there to be an iguana than a python? What means do languages have to indicate the strength of a conclusion from a piece of evidence? How do they compare possibilities? How do they talk about degrees of possibilities? Most importantly, how does grammar dip into this jungle of concepts and map them onto hierarchically structured sequences of words that we can use to reason about uncertainty in science and in daily life?”

The investigation of linguistic meaning: in the armchair, in the field, and in the lab

July 20 to 31, 2015, Berlin, Germany, Wissenschaftskolleg and ZAS
July 18 to 29, 2016, National Humanities Center, Research Triangle Park, North Carolina

sias-2016-gallery

The Summer Institute attracted 20 early-career researchers (PhD between 2009 and 2015) with one of three areas of specialization: (a) Theoretical Linguistics, especially Semantics and its interfaces, (b) Cognitive Psychology and Cognitive Neuroscience, and (c) Linguistic Fieldwork. One goal of the Summer Institute was interdisciplinary team building, resulting in joint projects, presentations, and publications. A second goal was capacity building, especially exposure to methods in the neighboring fields.

Conveners
Angelika Kratzer, Professor of Linguistics, University of Massachusetts at Amherst.
Manfred Krifka, Professor of General Linguistics at Humboldt Universität Berlin and Director of the Zentrum für Sprachwissenschaft, Berlin (ZAS).

Guest lecturers
Emmanuel Chemla, Research Scientist (CNRS), Laboratoire de Sciences Cognitives et Psycholinguistique, École Normale Supérieure, Paris.
Lisa Matthewson, Professor, Department of Linguistics, University of British Columbia.
Jesse Snedeker, Professor, Department of Psychology, Harvard University.
Malte Zimmermann, Professor of Semantics and Theory of Grammar, Universität Potsdam.

The program
SIAS Summer Institutes are designed to support the development of scholarly networks and collaborative projects among young scholars from the United States and Europe. The institutes are open to scholars who have received a Ph.D. within the past five years and Ph.D. candidates who are now studying or teaching at a European or American institution of higher education. Each institute accommodates twenty participants and is built around two summer workshops, one held in the United States and another in Europe in consecutive years. More info about SIAS Summer Institutes. And more info about the 2015 – 2016 SIAS Summer Institute

Sponsors and administration
SIAS Summer Institute are sponsored by SIAS (Some Institutes for Advanced Study). In the United States the 2015/16 Summer Institute was administered by the National Humanities Center. In Europe it was administered by the Wissenschaftskolleg zu Berlin. The program was made possible by grants from the Andrew W. Mellon Foundation and the Alexander von Humboldt Foundation.

Meaning composition in the brain

Source: Evelina Fedorenko, Terri L. Scott, Peter Brunnerd, William G. Coon, Brianna Pritchett, Gerwin Schalk, and Nancy Kanwisher (2016): Neural correlate of the construction of sentence meaning. Proceedings of the National Academy of Sciences.

“How do circuits of neurons in your brain extract and hold the meaning of a sentence? To start to address this unanswered question, we measured neural activity from the surface of the human brain in patients being mapped out before neurosurgery, as they read sentences. In many electrodes, neural activity increased steadily over the course of the sentence, but the same was not found when participants read lists of words or pronounceable nonwords, or grammatical nonword strings (“Jabberwocky”). This build-up of neural activity appears to reflect neither word meaning nor syntax alone, but the representation of complex meanings.”

The brain dictionary

From Nature: “Where exactly are the words in your head? Scientists have created an interactive map showing which brain areas respond to hearing different words. The map reveals how language is spread throughout the cortex and across both hemispheres, showing groups of words clustered together by meaning. The beautiful interactive model allows us to explore the complex organisation of the enormous dictionaries in our heads.” Explore the brain model for yourself hereRead the paper here.

Interestingly, function words (or logical words) like and, or, if, no, may and any kind of inflectional elements are mostly missing from this map (an exception are, interestingly, numerals). Can we conclude from this finding that LANGUAGE is spread throughout the brain? Or are we seeing a system of concepts that can be expressed by certain content vocabulary items in English? This work is exciting, the visuals are mind-blowing, but is this a finding about language?

Brain scientists discover meaning composition

If there is something that unites linguists of all stripes it’s the recognition that, in some way or other, natural languages systematically represent the cognitive distinction between the agent and the theme or patient of an action. The distinction is known to play a crucial role in the representation of verb meanings and the way they semantically compose with the verb’s arguments. 

An article in the Harvard Gazette of October 5 reports that two neuroscientists at Harvard discovered that the brain represents agents and patients/themes of actions in two distinct adjacent regions. The experiments are beautiful. This would be an occasion for linguists to celebrate if it wasn’t for the fact that the finding is being reported as a discovery about how the brain builds new thoughts. I don’t know who is ultimately responsible for aggrandizing the nature of the finding. Already in the original article in the Proceedings of the National Academy of Sciences the reader is led to believe that what was found went beyond the mere neural localization of a well-established semantic distinction. At the beginning of the PNAS article, the significance of the reported findings is described as follows:

“The 18th-century Prussian philosopher Wilhelm von Humbolt famously noted that natural language makes “infinite use of finite means.” By this, he meant that language deploys a finite set of words to express an effectively infinite set of ideas. As the seat of both language and thought, the human brain must be capable of rapidly encoding the multitude of thoughts that a sentence could convey. How does this work? Here, we find evidence supporting a long-standing conjecture of cognitive science: that the human brain encodes the meanings of simple sentences much like a computer, with distinct neural populations representing answers to basic questions of meaning such as “Who did it?” and “To whom was it done?” “

The bulk of this paragraph  describes no less than the research program of modern linguistics. How can the National Academy of Sciences tolerate such a disconnect between two disciplines within the Cognitive Sciences? How is it possible that a paper in neuroscience describes as a sensational new finding something that amounts to no more than the localization in the brain of a well-known and well-researched semantic distinction? We’ll never make progress in Cognitive Science if we allow subdisciplines to ignore each other so as to increase the perceived importance of one’s own results. 

One quote (attributed to Steven Frankland) in the Harvard Gazette article may point to the source of the communication problem: “This [the systematic representation of agents and themes/patients, A.K.] has been a central theoretical discussion in cognitive science for a long time, and although it has seemed like a pretty good bet that the brain works this way, there’s been little direct empirical evidence for it.” This quote makes it appear as if the idea that the human mind systematically represents agents and themes/patients has had the mere status of a bet before the distinction could be actually localized in the brain. That the distinction is systematically represented in all languages of the world is not given the status of a fact in this quote – it doesn’t count as  “empirical evidence”. It’s like denying your pulse the status of a fact before we can localize the mechanisms that regulate it in the brain. 

Headlines are routinely used for false advertising in the Cognitive Sciences. Science should be immune to those practices. The reported finding concerns a tiny aspect of meaning composition of the most simple kind and it implies nothing about the exact mechanism of composition. There is no empirical basis for drawing grand conclusions about “how the brain builds new thoughts” or about the brain “architecture for encoding sentence meaning.” Exaggerated headlines should be on the list of unscholarly behaviors that journal editors might want to discourage. 

Sources: How the brain builds new thoughts (Harvard Gazette, October 8) and the September 15) issue of the Proceedings of the National Academy of Sciences: An architecture for encoding sentence meaning in left mid-superior temporal cortex.  

Beth Stevens: Learning by cutting connections?

Source: MacArthur Foundation

We are born with an excess of synaptic connections. Through a normal developmental process called “pruning”, some of those synaptic connections get cut. Which ones? How are synaptic connections selected for elimination? What does all of this mean for our theories of how linguistic patterns and structures are acquired by children? 

“Beth Stevens is a neuroscientist whose research on microglial cells is prompting a significant shift in thinking about neuron communication in the healthy brain and the origins of adult neurological diseases. Until recently, it was believed that the primary function of microglia was immunological; they protected the brain by reducing inflammation and removing foreign bodies.

Stevens identified an additional, yet critical, role: the microglia are responsible for the “pruning” or removal of synaptic cells during brain development. Synapses form the connections, or means of communication, between nerve cells, and these pathways are the basis for all functions or jobs the brain performs. Using a novel model system that allows direct visualization of synapse pruning at various stages of brain development, Stevens demonstrated that the microglia’s pruning depends on the level of activity of neural pathways. She identified immune proteins called complement that “tag” (or bind) excess synapses with an “eat me” signal in the healthy developing brain. Through a process of phagocytosis, the microglia engulf or “eat” the synapses identified for elimination. This pruning optimizes the brain’s synaptic arrangements, ensuring that it has the most efficient wiring.”

Related articles: Microglia: New Roles for the Synaptic Stripper (Neuron) and Phagocytic cells: sculpting synaptic circuits in the developing nervous system

The neural code for belief ascriptions

koster-Hale

Jorie Koster-Hale

When I read in the February 28, 2014 issue of Science that the neural code for phonetic features had been discovered, I felt that this was a major breakthrough. I also felt, though, that marketing this finding as the discovery of “the neural code that makes us human” was greatly exaggerated. I wished at the time that similar discoveries could be made about semantic features. I did not know then that Jorie Koster-Hale’s dissertation (September 2014) would report precisely those kinds of findings. I think her discoveries are game-changing for all of us who are interested in the semantics and acquisition of attitude ascriptions. The prospect of being able to find the neural code for at least some of the features appearing in the semantic representations of belief ascriptions may profoundly change the way we do semantics. We will be held to much higher standards. For the first time, we may be able to do what – even a few weeks ago – I thought we weren’t yet able to do: relate some of the crucial features of semantic representations to a neural level of representation.

Among other things, Koster-Hale found that epistemic properties of other people’s beliefs are represented via response patterns of neural populations in canonical Theory of Mind regions. These properties relate to the kind of evidence that ground a belief: whether it was good evidence or not, or whether it was visual or auditory evidence. These properties trigger a major distinction in the class of attitude verbs: verbs in the believe family (believe thatsuspect thatconjecture that) can be used to report false beliefs, while verbs in the know family (know thatdiscover thatreveal that, hear that, see that) cannot – those verbs can only describe attitudes that are properly connected to reality. The very same epistemic properties are also known to be grammaticalized in evidential paradigms across languages, as Koster-Hale points out (see Aikhenvald 2004).

Related article in Cognition: “Thinking about seeing: Perceptual sources of knowledge are encoded in the theory of mind brain regions of sighted and blind adults.”

Moving beyond Big Data

Source: University of Maryland Language Science Center

“The National Science Foundation has announced that the University of Maryland’s Language Science Center (LSC) will (again!) receive a $3M grant for innovative research and graduate training, this time as part of the first cohort of awards made through its new NSF Research Traineeship (NRT) program. This 5-year award will support a model of interdisciplinary graduate training that prepares students to be adaptable scientists in multiple settings and career paths. The project will connect research on humans and machines, via a focus on how to succeed when Big Data is not available. The project is led by faculty and students from 10 departments across the entire university.”

“I’m excited by the research theme, which takes a “Beyond Big Data” approach,” says Colin Phillips, program PI and LSC Director. “We’re interested in how humans and machines can learn more efficiently from ‘multi-scale data’. Everybody’s talking these days about Big Data, but the current frontier in language science involves how to do more with less, you could call it ‘medium data’ or ‘small data’. It’s important for building better language technology, and it’s important for improving language learning outcomes in children and adults. Current language technologies like Google Translate and Apple’s Siri rely on a Big Data approach that stores billions of utterances. But that approach won’t generalize to the vast majority of the world’s 7000 languages. And human children easily outperform the best current technology, though they learn language from far less data. Child brains somehow learn the language around them more efficiently. But we all know that learning a new language as an adult is much harder. And we’re learning more and more about how children who experience ‘language poverty’, growing up with smaller amounts of quality language interaction, face negative consequences that last a lifetime. We want to understand why some learners fare better than others.” A key venue for exploring the program’s research goals will be yearly “Summer Camps”, intensive research-only workshops that will bring together students and faculty from UMD and beyond.”
Colin Phillips video.

The Latin American School for Education, Cognitive and Neural Sciences (LASchool)

logo1

LA School

From the website of the 5th instantiation of the LASchool: LASchool is a meeting that brings together students and faculties from all over the world in Latin America “to build new bridges between Education, Cognitive and Neural Sciences. Each year, LASchool’s participants work together for two weeks to generate project proposals potentially relevant for the development, design and implementation of effective science-based educational practices. LASchool series have been inspired by the ideas raised in The Santiago Declaration, in 2007.”

“Previous LASchools took place in Atacama, Chile (2011), Patagonia, Argentina (2012), Bahia, Brazil (2013), and Punta del Este, Uruguay (2014). All these experiences have brought together more than 150 researchers and 200 students in a continuing effort to promote the scientific work at the interface between Education and Science.”

This 5th version of the LASchool is organized by the Cognitive Neuroscience Laboratory at Pontificia Universidad Católica de Chile and will take place in San Pedro de Atacama. The LASchool will develop several issues such as the transition from informal to formal education, how brain systems change through development and education, and how social programs may impact education.”

The man who tried to redeem the world with logic

pitts

Source: Nautilus

From Nautilus:

“Though they started at opposite ends of the socioeconomic spectrum, McCulloch and Pitts were destined to live, work, and die together. Along the way, they would create the first mechanistic theory of the mind, the first computational approach to neuroscience, the logical design of modern computers, and the pillars of artificial intelligence. But this is more than a story about a fruitful research collaboration. It is also about the bonds of friendship, the fragility of the mind, and the limits of logic’s ability to redeem a messy and imperfect world.”

“The moment they spoke, they realized they shared a hero in common: Gottfried Leibniz. The 17th-century philosopher had attempted to create an alphabet of human thought, each letter of which represented a concept and could be combined and manipulated according to a set of logical rules to compute all knowledge—a vision that promised to transform the imperfect outside world into the rational sanctuary of a library.”