Search

Due to the compelling improvements brought by BERT, many recent representation models adopted the Transformer architecture as their main building block, consequently inheriting the wordpiece tokenization system despite it not being intrinsically linked to the notion of Transformers. While this system is thought to achieve a good balance between the flexibility of characters and the efficiency of full words, using predefined wordpiece vocabularies from the general domain is not always suitable, especially when building models for specialized domains (e.g., the medical domain). Moreover, adopting a wordpiece tokenization shifts the focus from the word level to the subword level, making the models conceptually more complex and arguably less convenient in practice. For these reasons, we propose CharacterBERT, a new variant of BERT that drops the wordpiece system altogether and uses a Character-CNN module instead to represent entire words by consulting their characters. We show that this new model improves the performance of BERT on a variety of medical domain tasks while at the same time producing robust, word-level and open-vocabulary representations., Comment: 13 pages, 8 figures and 3 tables. Accepted at COLING 2020

Inelastic proton scattering experiments were performed at the Research Center for Nuclear Physics, Osaka, with a 295 MeV beam covering laboratory angles 0{\deg}-6{\deg} and excitation energies 6-22 MeV. Cross sections due to E1 and M1 excitations were extracted with a multipole decomposition analysis and then converted to reduced transition probabilities with the "virtual photon method" for E1 and the "unit cross section method" for M1 excitations, respectively. Including a theory-aided correction for the high excitation energy region not covered experimentally, the electric dipole polarizability was determined from the E1 strength distributions. Total photoabsorption cross sections derived from the E1 and M1 strength distributions show significant differences compared to those from previous ($\gamma$,xn) experiments in the energy region of the isocvector giant dipole resonance (IVGDR). The widths of the IVGDR deduced from the present data with a Lorentz parameterization show an approximately constant value of about 4.5 MeV in contrast to the large variations between isotopes observed in previous work. The IVGDR centroid energies are in good correspondence to expectations from systematics of their mass dependence. Furthermore, a study of the dependence of the IVGDR energies on bulk matter properties is presented. The E1 strengths below neutron threshold show fair agreement with results from ($\gamma$,$\gamma$') experiments on 112,116,120,124Sn in the energy region between 6 and 7 MeV. At higher excitation energies large differences are observed pointing to a different nature of the excited states with small ground state branching ratios. The isovector spin-M1 strengths exhibit a broad distribution between 6 and 12 MeV in all studied nuclei., Comment: 32 pages, 23 figures, submitted to Phys. Rev. C

The dipole polarizability of stable even-mass tin isotopes 112,114,116,118,120,124 was extracted from inelastic proton scattering experiments at 295 MeV under very forward angles performed at RCNP. Predictions from energy density functionals cannot account for the present data and the polarizability of 208Pb simultaneously. The evolution of the polarizabilities in neighboring isotopes indicates a kink at 120Sn while all model results show a nearly linear increase with mass number after inclusion of pairing corrections., Comment: 10 pages, 6 figures, submitted to Phys. Lett. B

We explore the utilities of explicit negative examples in training neural language models. Negative examples here are incorrect words in a sentence, such as "barks" in "*The dogs barks". Neural language models are commonly trained only on positive examples, a set of sentences in the training data, but recent studies suggest that the models trained in this way are not capable of robustly handling complex syntactic constructions, such as long-distance agreement. In this paper, using English data, we first demonstrate that appropriately using negative examples about particular constructions (e.g., subject-verb agreement) will boost the model's robustness on them, with a negligible loss of perplexity. The key to our success is an additional margin loss between the log-likelihoods of a correct word and an incorrect word. We then provide a detailed analysis of the trained models. One of our findings is the difficulty of object-relative clauses for RNNs. We find that even with our direct learning signals the models still suffer from resolving agreement across an object-relative clause. Augmentation of training sentences involving the constructions somewhat helps, but the accuracy still does not reach the level of subject-relative clauses. Although not directly cognitively appealing, our method can be a tool to analyze the true architectural limitation of neural models on challenging linguistic constructions., Comment: ACL 2020 camera ready (long paper); code is available at https://github.com/aistairc/lm_syntax_negative

Inelastic ${}^{6}$Li scattering at 100 MeV/u on ${}^{12}$C and ${}^{93}$Nb have been measured with the high-resolution magnetic spectrometer Grand Raiden. The magnetic-rigidity settings of the spectrometer covered excitation energies from 10 to 40 MeV and scattering angles in the range $0^\circ < \theta_{\text{lab.}}< 2^\circ$. The isoscalar giant monopole resonance was selectively excited in the present data. Measurements free of instrumental background and the very favorable resonance-to-continuum ratio of ${}^{6}$Li scattering allowed for precise determination of the $E0$ strengths in ${}^{12}$C and ${}^{93}$Nb. It was found that the monopole strength in ${}^{12}$C exhausts $52 \pm 3^\text{(stat.)} \pm 8 ^\text{(sys.)}$\% of the energy-weighted sum rule (EWSR), which is considerably higher than results from previous $\alpha$-scattering experiments. The monopole strength in ${}^{93}$Nb exhausts $92 \pm 4^\text{(stat.)} \pm 10 ^\text{(sys.)}$\% of the EWSR, and it is consistent with measurements of nuclei with mass number of $A\approx90$. Such comparison indicates that the isoscalar giant monopole resonance distributions in these nuclei are very similar, and no influence due to nuclear structure was observed., Comment: 6 Figures

We have succeeded in synthesizing electron-doped polycrystalline bulk samples of T'-La1.8Eu0.2CuO4-yFy (y = 0 - 0.15) by the fluorination of undoped (Ce-free) T'-La1.8Eu0.2CuO4 using NH4F. The magnetic susceptibility measurements have revealed that the superconducting transition temperature, Tc, increases with increasing y, exhibits the maximum of 23 K at y = 0.025, and decreases. The dome-like dependence of Tc on the doped carrier concentration in the T'-type (La,Eu)-based cuprates is explained in terms of the pairing mediated by spin fluctuations based on the d-p model calculation [K. Yamazaki et al., J. Phys.: Conf. Ser. 871, 012009 (2017)]., Comment: 14 pages, 5 figures

We report the achieving of depairing current limit along $c$-axis in Fe$_{1+y}$Te$_{1-x}$Se$_x$ single crystals. A series of crystals with $T_{\rm{c}}$ ranging from 8.6 K to 13.7 K (different amount of excess Fe, $y$) were fabricated into $c$-axis bridges with a square-micrometer cross-section. The critical current density, $J_{\rm{c}}$, was directly estimated from the transport current-voltage measurements. The transport $J_{\rm{c}}$ reaches a very large value, which is about one order of magnitude larger than the depinning $J_{\rm{c}}$, but comparable to the calculated depairing $J_{\rm{c}}$ $\sim$ 2 $\times$ 10$^6$ A/cm$^2$ at 0 K, based on the Ginzburg-Landau (GL) theory. The temperature dependence of the depairing $J_{\rm{c}}$ follows the GL-theory ($\propto$ (1-$T/T_{\rm{c}}$)$^{3/2}$) down to $\sim$ 0.83 $T_{\rm{c}}$, then increases with a reduced slope at low temperatures, which can be qualitatively described by the Kupriyanov-Lukichev theory. Our study provides a new route to understand the behavior of depairing $J_{\rm{c}}$ in iron-based superconductors in a wide temperature range., Comment: 8 pages, 6 figures

We have measured temperature and magnetic field dependences of the thermal conductivity along the c-axis, kc, and that along the [110] direction, k110, of CuB2O4 single crystals in zero field and magnetic fields along the c-axis and along the [110] direction. It has been found that the thermal conductivity is nearly isotropic and very large in zero field and that the thermal conductivity due to phonons is dominant in CuB2O4. The temperature and field dependences of kc and k110 have markedly changed at phase boundaries in the magnetic phase diagram, which has been understood to be due to the change of the mean free path of phonons caused by the change of the phonon-spin scattering rate at the phase boundaries. It has been concluded that thermal conductivity measurements are very effective for detecting magnetic phase boundaries., Comment: 18 pages, 6 figures

In the late stages of stellar core-collapse, prior to core bounce, electron captures on medium-heavy nuclei drive deleptonization and simulations require the use of accurate reaction rates. Nuclei with neutron number near $N=50$, just above atomic number $Z=28$, play an important role, but rates used in astrophysical simulations rely primarily on a relatively simple single-state approximation. In order to improve the accuracy of astrophysical simulations, experimental data are needed to test the electron-capture rates and to guide the development of better theoretical models. This work presents the results of the $^{86}$Kr($t$,$^{3}$He+$\gamma$) experiment at the NSCL, from which an upper limit for the Gamow-Teller strength up to an excitation energy in $^{86}$Br of 5 MeV is extracted. The derived upper limit for the electron-capture rate on $^{86}$Kr indicates that the rate estimated through the single-state approximation is too high and that rates based on Gamow-Teller strengths estimated in shell-model and QRPA calculations are more accurate. The QRPA calculations tested in this manner were used for estimating the electron capture rates for 78 isotopes near $N=50$ and above $Z=28$. The impact of using these new electron-capture rates in simulations of supernovae instead of the rates based on the single-state approximation is investigated, indicating a significant reduction in the deleptonization that affects multi-messenger signals, such as the emission of neutrinos and gravitational waves., Comment: 15 pages, 10 figures

We propose a data-to-text generation model with two modules, one for tracking and the other for text generation. Our tracking module selects and keeps track of salient information and memorizes which record has been mentioned. Our generation module generates a summary conditioned on the state of tracking module. Our model is considered to simulate the human-like writing process that gradually selects the information by determining the intermediate variables while writing the summary. In addition, we also explore the effectiveness of the writer information for generation. Experimental results show that our model outperforms existing models in all evaluation metrics even without writer information. Incorporating writer information further improves the performance, contributing to content planning and surface realization., Comment: ACL 2019

The Gamow-Teller strength distribution from ${}^{88}$Sr was extracted from a $(t,{}^{3}\text{He}+\gamma)$ experiment at 115 MeV/$u$ to constrain estimates for the electron-capture rates on nuclei around $N=50$, between and including $^{78}$Ni and $^{88}$Sr, which are important for the late evolution of core-collapse supernovae. The observed strength below an excitation energy of 8 MeV was consistent with zero and below 10 MeV amounted to $0.1\pm0.05$. Except for a very-weak transition that could come from the 2.231-MeV $1^{+}$ state, no $\gamma$ lines that could be associated with the decay of known $1^{+}$ states were identified. The derived electron-capture rate from the measured strength distribution is more than an order of magnitude smaller than rates based on the single-state approximation presently used in astrophysical simulations for most nuclei near $N=50$. Rates based on shell-model and quasiparticle random-phase approximation calculations that account for Pauli blocking and core-polarization effects provide better estimates than the single-state approximation, although a relatively strong transition to the first $1^{+}$ state in $^{88}$Rb is not observed in the data. Pauli unblocking effects due to high stellar temperatures could partially counter the low electron-capture rates. The new data serves as a zero-temperature benchmark for constraining models used to estimate such effects.

We propose a new domain adaptation method for Combinatory Categorial Grammar (CCG) parsing, based on the idea of automatic generation of CCG corpora exploiting cheaper resources of dependency trees. Our solution is conceptually simple, and not relying on a specific parser architecture, making it applicable to the current best-performing parsers. We conduct extensive parsing experiments with detailed discussion; on top of existing benchmark datasets on (1) biomedical texts and (2) question sentences, we create experimental datasets of (3) speech conversation and (4) math problems. When applied to the proposed method, an off-the-shelf CCG parser shows significant performance gains, improving from 90.7% to 96.6% on speech conversation, and from 88.5% to 96.8% on math problems., Comment: 11 pages, accepted as long paper to ACL 2019 Italy

We investigated the magnetism and superconductivity in as-sintered (AS) and oxidation annealed (OA) T*-type La$_{1-x/2}$Eu$_{1-x/2}$Sr$_x$CuO$_4$ (LESCO) with 0.14 $\leq x \leq$ 0.28 by the first comprehensive muon spin rotation/relaxation ($\mu$SR), magnetic susceptibility, and electrical resistivity measurements. In OA superconducting samples, no evidence of magnetic order was observed, whereas AS semiconducting samples exhibited evidence of a disordered magnetic state in the measured temperature range between $\sim$4 K and $\sim$8 K. Therefore, the ground state in LESCO drastically varies with oxidation annealing and the magnetic phase competitively exists with the superconducting (SC) phase. The magnetic phase in the AS LESCO is quite robust against Sr doping, while the SC phase degrades with increasing $x$. A monotonous decrease of the SC transition temperature from 24.5 K in $x$ = 0.14 to 9.0 K in $x$ = 0.28 suggests the disappearance of the SC phase at $x$ $\sim$ 0.34. Furthermore, we clarified the simultaneous development of (quasi) static magnetism and the electrical resistivity at a low temperature in AS samples, suggesting the inducement of magnetism by the suppression of carrier mobility. The variation in magnetism due to annealing is discussed from a viewpoint of structural defects, which was previously reported from neutron diffraction measurements.

Imaging and analysis of many single cells hold great potential in our understanding of heterogeneous and complex life systems and in enabling biomedical applications. We here introduce a recently realized image-free "imaging" cytometry technology, which we call ghost cytometry. While a compressive ghost imaging technique utilizing object's motion relative to a projected static light pattern allows recovery of their images, a key of this ghost cytometry is to achieve ultrafast cell classification by directly applying machine learning methods to the compressive imaging signals in a temporal domain. We show the applicability of our method in the analysis of flowing objects based on the reconstructed images as well as in that based on the imaging waveform without image production.

Imaging flow cytometry shows significant potential for increasing our understanding of heterogeneous and complex life systems and is useful for biomedical applications. Ghost cytometry is a recently proposed approach for directly analyzing compressively measured signals, thereby relieving the computational bottleneck observed in high-throughput cytometry based on morphological information. While this image-free approach could distinguish different cell types using the same fluorescence staining method, further strict controls are sometimes required to clearly demonstrate that the classification is based on detailed morphologic analysis. In this study, we show that ghost cytometry can be used to classify cell populations of the same type but with different fluorescence distributions in space, supporting the strength of our image-free approach for morphologic cell analysis.

It is known that for the natural algebraic torus actions on the Grassmannians, the closures of torus orbits are toric varieties, and that these toric varieties are smooth if and only if the corresponding matroid polytopes are simple. We prove that simple matroid polytopes are products of simplices and smooth torus orbit closures in the Grassmannians are products of complex projective spaces. Moreover, it turns out that the smooth torus orbit closures are uniquely determined by the corresponding simple matroid polytopes., Comment: 12 pages, 9 figures