Recent works have shown that language models (LM) capture different types of knowledge regarding facts or common sense. However, because no model is perfect, they still fail to provide appropriate answers in many cases. In this paper, we ask the question, “How can we know when language models know, with confidence, the answer to a particular query?” We examine this question from the point of view of calibration, the property of a probabilistic model’s predicted probabilities actually being well correlated with the probabilities of correctness. We examine three strong generative models—T5, BART, and GPT-2—and study whether their probabilities on QA tasks are well calibrated, finding the answer is a relatively emphatic no. We then examine methods to calibrate such models to make their confidence scores correlate better with the likelihood of correctness through fine-tuning, post-hoc probability modification, or adjustment of the predicted outputs or inputs. Experiments on a diverse range of datasets demonstrate the effectiveness of our methods. We also perform analysis to study the strengths and limitations of these methods, shedding light on further improvements that may be made in methods for calibrating LMs. We have released the code at https://github.com/jzbjyb/lm-calibration.

Nanowhiskers of crystalline polysaccharides, including cellulose and chitin, have recently been attracting much attention, especially as rigid filler components of novel nanocomposite materials. Good dispersion of these nanowhiskers in aqueous suspensions, which are usually prepared by appropriate acid hydrolysis of native polysaccharide raw materials, has been examined based on two types of stabilization mechanisms, i.e., electrostatic and steric repulsions between nanowhiskers. In this review, recent results on electrostatic and steric stabilizations of nanowhiskers are introduced, together with brief and basic descriptions of nanowhiskers and these stabilization mechanisms. Some of the remarkable characteristics of nanowhiskers, such as viscosity behaviors, liquid-crystal formation, and nanocomposite formation, which are related to useful applications of these stable nanowhiskers, are also summarized.

A novel “sliding graft copolymer” (SGC), in which many linear poly-ε-caprolactone (PCL) side chains are bound to cyclodextrin rings of a polyrotaxane, was prepared by ring-opening polymerization of ε-caprolactone initiated by hydroxyl groups of the polyrotaxane. An amorphous, flexible, and sufficiently tough elastomer film was prepared by crosslinking the obtained SGC—a supramolecule possessing a number of mobile side chains—with hexamethylene diisocyanate (HMDI).

Nanocomposite gels containing chitosan as a network polymer and chitin nanowhiskers as reinforcing fillers were prepared by cross-linking with the blocked isocyanate hexamethylene-1,6-di-(aminocarboxysulfonate) (HDS). Young’s modulus and stress at break of the nanocomposite gels increased to maximum values of 169 and 135 kPa, respectively, with increase in whisker content up to 13.3%, whereas the degree of swelling decreased from 357 to 33.0.

A sterically stabilized aqueous suspension of “charge-free” cellulose nanowhiskers was prepared by hydrochloric acid hydrolysis of cotton powders and subsequent surface grafting of monomethoxy poly(ethylene glycol) (mPEG). The preparation scheme included carboxylation of the terminal hydroxyl groups in mPEG via oxidation with silica gel particles carrying 2,2,6,6-tetramethyl-1-pyperidinyloxyl (TEMPO) moieties and subsequent esterification between terminal carboxyls in mPEG and surface hydroxyl groups of cellulose nanowhiskers, mediated by 1,1'-carbonyldiimidazole (CDI) in dimethyl sulfoxide or dimethylacetamide. Some of the prepared PEG-grafted samples showed remarkable flow birefringence and enhanced stability after 24 h, even in 0.1 M NaCl, suggesting successful steric stabilization by efficient mPEG grafting. Actual PEG grafting via ester linkages was confirmed by attenuated total reflectance-Fourier transform infrared spectrometry. In a typical example, the amount of grafted mPEG was estimated as ca. 0.3 g/g cellulose by two measurements, i.e., weight increase after grafting and weight loss after alkali cleavage of ester linkages. Transmission electron microscopy indicated unchanged nanowhisker morphology after mPEG grafting.