Provided proper attribution is provided, Google hereby grants permission to reproduce the tables and figures in this paper solely for use in journalistic or scholarly works.

---

適切な帰属表示が行われる場合、Googleは本論文の表および図を、ジャーナリスティックまたは学術的な著作物での使用に限り複製することを許可します。

---

# Attention Is All You Need

**Ashish Vaswani**$^{*}$
Google Brain
avaswani@google.com

**Llion Jones**$^{*}$
Google Research
llion@google.com

---

**アシッシュ・ヴァスワニ**$^{*}$
Google Brain
avaswani@google.com

**リアン・ジョーンズ**$^{*}$
Google Research
llion@google.com

---

Noam Shazeer$^*$
Google Brain
noam@google.com

**Aidan N. Gomez$^{*}$ $^{\dagger}$**
University of Toronto
aidan@cs.toronto.edu

---

**Aidan N. Gomez$^{*}$ $^{\dagger}$**
トロント大学
aidan@cs.toronto.edu

---

\textbf{Niki Parmar}\*

Google Research

nikip@google.com

---

\textbf{ニキ・パルマール}\*

Google Research

nikip@google.com

---

Illia Polosukhin$^{*\ \ddagger}$

illia.polosukhin@gmail.com

---

イリア・ポロスクヒン$^{*\ \ddagger}$

illia.polosukhin@gmail.com

---

## Abstract

---

## 要約

---

Jakob Uszkoreit$^{\ast}$
Google Research
usz@google.com

Łukasz Kaiser*
Google Brain
lukaszkaiser@google.com

The dominant sequence transduction models are based on complex recurrent or convolutional neural networks that include an encoder and a decoder. The best performing models also connect the encoder and decoder through an attention mechanism. We propose a new simple network architecture, the Transformer, based solely on attention mechanisms, dispensing with recurrence and convolutions entirely. Experiments on two machine translation tasks show these models to be superior in quality while being more parallelizable and requiring significantly less time to train. Our model achieves 28.4 BLEU on the WMT 2014 English-to-German translation task, improving over the existing best results, including ensembles, by over 2 BLEU. On the WMT 2014 English-to-French translation task, our model establishes a new single-model state-of-the-art BLEU score of 41.8 after training for 3.5 days on eight GPUs, a small fraction of the training costs of the best models from the literature. We show that the Transformer generalizes well to other tasks by applying it successfully to English constituency parsing both with large and limited training data.

---

支配的なシーケンス変換モデルは、エンコーダとデコーダを含む複雑な再帰型または畳み込みニューラルネットワークに基づいている。最も性能の高いモデルは、さらにエンコーダとデコーダを注意機構（アテンションメカニズム）で接続している。我々は、再帰や畳み込みを完全に排除し、注意機構のみに基づく新しい単純なネットワークアーキテクチャであるTransformerを提案する。2つの機械翻訳タスクにおける実験では、これらのモデルが品質において優れており、より並列化可能で、訓練に要する時間が大幅に短いことを示した。本モデルは、WMT 2014の英語からドイツ語への翻訳タスクで28.4 BLEUを達成し、アンサンブルを含む既存の最良結果を2 BLEU以上上回った。また、WMT 2014の英語からフランス語への翻訳タスクでは、8つのGPU上で3.5日間訓練した後、41.8という単一モデルとしての新たなstate-of-the-art BLEUスコアを樹立し、これは文献にある最良モデルの訓練コストのごく一部であった。さらに、大規模および限定的な訓練データを用いた英語の統語木構文解析に本モデルを成功적으로適用することで、Transformerが他のタスクにもよく汎化することを示した。

---

arXiv:1706.03762v7 [cs.CL] 2 Aug 2023

$^*$Equal contribution. Listing order is random. Jakob proposed replacing RNNs with self-attention and started the effort to evaluate this idea. Ashish, with Illia, designed and implemented the first Transformer models and has been crucially involved in every aspect of this work. Noam proposed scaled dot-product attention, multi-head attention and the parameter-free position representation and became the other person involved in nearly every detail. Niki designed, implemented, tuned and evaluated countless model variants in our original codebase and tensor2tensor. Llion also experimented with novel model variants, was responsible for our initial codebase, and efficient inference and visualizations. Lukasz and Aidan spent countless long days designing various parts of and implementing tensor2tensor, replacing our earlier codebase, greatly improving results and massively accelerating our research.

$^\dagger$Work performed while at Google Brain.

$^\ddagger$Work performed while at Google Research.

---

$^*$同等貢献。記載順はランダムです。JakobはRNNを自己注意機構に置き換えることを提案し、このアイデアの評価に取り組み始めました。AshishはIlliaと共に最初のTransformerモデルの設計と実装を行い、この作業のあらゆる側面に重要な役割を果たしました。Noamはスケーリングされたドット積アテンション、マルチヘッド・アテンション、パラメータなしの位置表現を提案し、ほぼすべての詳細に関与したもう一人の主要な人物となりました。Nikiは、元のコードベースおよびtensor2tensorにおいて、無数のモデル変種の設計、実装、チューニング、評価を行いました。Llionもまた、新規なモデル変種の実験を行い、初期のコードベース、効率的な推論、可視化を担当しました。LukaszとAidanは、tensor2tensorの様々な部分の設計と実装に多数の長時間を費やし、以前のコードベースを置き換えることで結果を大幅に改善し、研究を飛躍的に加速させました。

$^\dagger$Google Brain在籍中に実施した研究。

$^\ddagger$Google Research在籍中に実施した研究。

---

## 1 Introduction

---

## 1 導入

---

Recurrent neural networks, long short-term memory [13] and gated recurrent [7] neural networks in particular, have been firmly established as state of the art approaches in sequence modeling and transduction problems such as language modeling and machine translation [35, 2, 5]. Numerous efforts have since continued to push the boundaries of recurrent language models and encoder-decoder architectures [38, 24, 15].

Recurrent models typically factor computation along the symbol positions of the input and output sequences. Aligning the positions to steps in computation time, they generate a sequence of hidden states $h_t$, as a function of the previous hidden state $h_{t-1}$ and the input for position $t$. This inherently sequential nature precludes parallelization within training examples, which becomes critical at longer sequence lengths, as memory constraints limit batching across examples. Recent work has achieved significant improvements in computational efficiency through factorization tricks [21] and conditional computation [32], while also improving model performance in case of the latter. The fundamental constraint of sequential computation, however, remains.

Attention mechanisms have become an integral part of compelling sequence modeling and transduction models in various tasks, allowing modeling of dependencies without regard to their distance in the input or output sequences [2, 19]. In all but a few cases [27], however, such attention mechanisms are used in conjunction with a recurrent network.

In this work we propose the Transformer, a model architecture eschewing recurrence and instead relying entirely on an attention mechanism to draw global dependencies between input and output. The Transformer allows for significantly more parallelization and can reach a new state of the art in translation quality after being trained for as little as twelve hours on eight P100 GPUs.

---

再帰型ニューラルネットワーク、特にロングショートタームメモリ[13]やゲート付き再帰型ニューラルネットワークは、言語モデリングや機械翻訳といった系列モデル化および変換問題において最先端のアプローチとして確固たる地位を築いています[35, 2, 5]。その後、再帰型言語モデルやエンコーダ・デコーダアーキテクチャの限界を引き上げるための取り組みが数多く行われてきました[38, 24, 15]。

再帰型モデルは通常、入力および出力系列の記号位置に沿って計算を分割します。計算時間のステップに位置を対応させることで、以前の隠れ状態 $h_{t-1}$ と位置 $t$ の入力を関数として用い、隠れ状態の系列 $h_t$ を生成します。この本質的に逐次的な性質により、トレーニング例内での並列化が不可能となり、これはより長い系列長において重要になります。なぜなら、メモリ制約が例間のバッチ処理を制限するためです。最近の研究では、因子分解トリック[21]や条件付き計算[32]を通じて計算効率を大幅に向上させるとともに、後者の場合にはモデル性能も改善しています。しかし、逐次計算という根本的な制約は依然として残っています。

注意機構（Attention mechanisms）は、様々なタスクにおける魅力的な系列モデル化および変換モデルの不可欠な部分となっており、入力または出力系列内での距離に関係なく依存関係をモデル化することを可能にします[2, 19]。ただし、ごく一部のケース[27]を除き、これらの注意機構は再帰型ネットワークと組み合わせて使用されます。

本研究では、再帰を避け、代わりに注意機構のみを用いて入力と出力の間にグローバルな依存関係を引き出すモデルアーキテクチャであるTransformerを提案します。Transformerは大幅な並列化を可能にし、8つのP100 GPU上でわずか12時間トレーニングするだけで、翻訳品質において新たな最先端を達成することができます。

---

## 2 Background

---

## 2 背景

---

The goal of reducing sequential computation also forms the foundation of the Extended Neural GPU [16], ByteNet [18] and ConvS2S [9], all of which use convolutional neural networks as basic building block, computing hidden representations in parallel for all input and output positions. In these models, the number of operations required to relate signals from two arbitrary input or output positions grows in the distance between positions, linearly for ConvS2S and logarithmically for ByteNet. This makes it more difficult to learn dependencies between distant positions [12]. In the Transformer this is reduced to a constant number of operations, albeit at the cost of reduced effective resolution due to averaging attention-weighted positions, an effect we counteract with Multi-Head Attention as described in section 3.2.

Self-attention, sometimes called intra-attention is an attention mechanism relating different positions of a single sequence in order to compute a representation of the sequence. Self-attention has been used successfully in a variety of tasks including reading comprehension, abstractive summarization, textual entailment and learning task-independent sentence representations [4, 27, 28, 22].

End-to-end memory networks are based on a recurrent attention mechanism instead of sequence-aligned recurrence and have been shown to perform well on simple-language question answering and language modeling tasks [34].

To the best of our knowledge, however, the Transformer is the first transduction model relying entirely on self-attention to compute representations of its input and output without using sequence-aligned RNNs or convolution. In the following sections, we will describe the Transformer, motivate self-attention and discuss its advantages over models such as [17, 18] and [9].

---

逐次計算を削減するという目標は、拡張ニューラルGPU [16]、ByteNet [18]、ConvS2S [9] の基盤にもなっており、これらはすべて畳み込みニューラルネットワークを基本構成要素として用い、すべての入力および出力位置に対して隠れ表現を並列に計算する。これらのモデルでは、2つの任意の入力または出力位置間の信号を関連付けるために必要な操作数は、位置間の距離とともに増加し、ConvS2Sでは線形に、ByteNetでは対数的に増加する。これは、遠く離れた位置間の依存関係を学習することをより困難にする [12]。Transformerでは、アテンション重み付けされた位置の平均化により有効解像度が低下するという代償を払うものの、この問題が一定数の操作数にまで削減される。この効果については、3.2節で説明するマルチヘッド・アテンションによって相殺する。

自己アテンション（intra-attentionとも呼ばれる）は、単一シーケンス内の異なる位置同士を関連付けて、そのシーケンスの表現を計算するためのアテンション機構である。自己アテンションは、読解、要約、テキストの含意関係、タスク非依存の文表現の学習など、さまざまなタスクで成功的に使用されてきた [4, 27, 28, 22]。

エンドツーエンドのメモリネットワークは、シーケンス整列型のリカレンスではなく再帰的アテンション機構に基づいており、単純な言語による質問応答や言語モデリングのタスクで良好な性能を示すことが示されている [34]。

しかしながら、我々の知る限り、Transformerは初めて、シーケンス整列型のRNNや畳み込みを使用せずに、自己アテンションのみを用いて入出力の表現を計算するトランスダクションモデルである。次の節では、Transformerについて説明し、自己アテンションの動機づけを行い、[17, 18] および [9] などのモデルに対する利点を議論する。

---

## 3 Model Architecture

---

## 3 モデルアーキテクチャ

---

Most competitive neural sequence transduction models have an encoder-decoder structure [5, 2, 35]. Here, the encoder maps an input sequence of symbol representations $(x_1, ..., x_n)$ to a sequence of continuous representations $\mathbf{z} = (z_1, ..., z_n)$. Given $\mathbf{z}$, the decoder then generates an output sequence $(y_1, ..., y_m)$ of symbols one element at a time. At each step the model is auto-regressive [10], consuming the previously generated symbols as additional input when generating the next.

---

最も競争力のあるニューラル系列変換モデルはエンコーダ・デコーダ構造を持っている [5, 2, 35]。ここで、エンコーダは記号表現の入力系列 $(x_1, ..., x_n)$ を連続表現の系列 $\mathbf{z} = (z_1, ..., z_n)$ にマッピングする。$\mathbf{z}$ が与えられたとき、デコーダは次に生成される記号を1つずつ消費しながら、出力系列 $(y_1, ..., y_m)$ を逐次生成する。各ステップにおいて、モデルは自己回帰的 [10] であり、次の記号を生成する際に、以前に生成された記号を追加の入力として使用する。

---

<img src="images/3-1.png" style="zoom:70%; display: block; margin: 0 auto;" />

Figure 1: The Transformer - model architecture.

---

図1: Transformer - モデルのアーキテクチャ。

---

The Transformer follows this overall architecture using stacked self-attention and point-wise, fully connected layers for both the encoder and decoder, shown in the left and right halves of Figure 1, respectively.

---

Transformerは、エンコーダとデコーダの両方でスタックされた自己注意機構とポイントワイズな全結合層を用いたこの全体アーキテクチャを採用しており、それぞれ図1の左側と右側に示されています。

---

## 3.1 Encoder and Decoder Stacks

---

## 3.1 エンコーダとデコーダのスタック

---

**Encoder:** The encoder is composed of a stack of $N = 6$ identical layers. Each layer has two sub-layers. The first is a multi-head self-attention mechanism, and the second is a simple, position-wise fully connected feed-forward network. We employ a residual connection [11] around each of the two sub-layers, followed by layer normalization [1]. That is, the output of each sub-layer is $\text{LayerNorm}(x + \text{Sublayer}(x))$, where $\text{Sublayer}(x)$ is the function implemented by the sub-layer itself. To facilitate these residual connections, all sub-layers in the model, as well as the embedding layers, produce outputs of dimension $d_{\text{model}} = 512$.

**Decoder:** The decoder is also composed of a stack of $N = 6$ identical layers. In addition to the two sub-layers in each encoder layer, the decoder inserts a third sub-layer, which performs multi-head attention over the output of the encoder stack. Similar to the encoder, we employ residual connections around each of the sub-layers, followed by layer normalization. We also modify the self-attention sub-layer in the decoder stack to prevent positions from attending to subsequent positions. This masking, combined with fact that the output embeddings are offset by one position, ensures that the predictions for position $i$ can depend only on the known outputs at positions less than $i$.

---

**エンコーダ:** エンコーダは、$N = 6$ 層の同一なレイヤーのスタックで構成されています。各レイヤーには2つのサブレイヤーがあります。1つ目はマルチヘッド自己注意機構であり、2つ目は単純な位置ごとの全結合フィードフォワードネットワークです。私たちは、各サブレイヤーの周りに残差接続 [11] を採用し、その後にレイヤー正規化 [1] を行います。つまり、各サブレイヤーの出力は $\text{LayerNorm}(x + \text{Sublayer}(x))$ となり、ここで $\text{Sublayer}(x)$ はサブレイヤー自体が実装する関数です。これらの残差接続を容易にするため、モデル内のすべてのサブレイヤーおよび埋め込み層は、次元 $d_{\text{model}} = 512$ の出力を生成します。

**デコーダ:** デコーダもまた、$N = 6$ 層の同一なレイヤーのスタックで構成されています。各エンコーダレイヤーの2つのサブレイヤーに加え、デコーダは3番目のサブレイヤーを挿入し、これはエンコーダスタックの出力に対してマルチヘッド注意を行います。エンコーダと同様に、各サブレイヤーの周りに残差接続を採用し、その後にレイヤー正規化を行います。また、デコーダスタック内の自己注意サブレイヤーを変更し、ある位置がそれ以降の位置に注意を向けるのを防ぎます。このマスクと、出力埋め込みが1つずれているという事実を組み合わせることで、位置 $i$ に対する予測が位置 $i$ より小さい既知の出力のみ依存するように保証されます。

---

## 3.2 Attention

---

## 3.2 アテンション

---

An attention function can be described as mapping a query and a set of key-value pairs to an output, where the query, keys, values, and output are all vectors. The output is computed as a weighted sum

---

アテンション関数は、クエリとキー・バリューのペアのセットを出力にマッピングするものとして記述できる。ここで、クエリ、キー、バリュー、および出力はすべてベクトルである。出力は重み付き和として計算される

---

<img src="images/4-3.png" style="zoom:70%; display: block; margin: 0 auto;" />

Figure 2: (left) Scaled Dot-Product Attention. (right) Multi-Head Attention consists of several attention layers running in parallel.

---

図2: (左) スケーリングされたドット積アテンション。(右) マルチヘッドアテンションは、複数のアテンション層が並列に動作する構成です。

---

of the values, where the weight assigned to each value is computed by a compatibility function of the query with the corresponding key.

---

の値であり、各値に割り当てられる重みは、クエリと対応するキーとの互換性関数によって計算される。

---

## 3.2.1 Scaled Dot-Product Attention

---

## 3.2.1 スケーリングされたドット積アテンション

---

We call our particular attention "Scaled Dot-Product Attention" (Figure 2). The input consists of queries and keys of dimension $d_k$, and values of dimension $d_v$. We compute the dot products of the query with all keys, divide each by $\sqrt{d_k}$, and apply a softmax function to obtain the weights on the values.

In practice, we compute the attention function on a set of queries simultaneously, packed together into a matrix $Q$. The keys and values are also packed together into matrices $K$ and $V$. We compute the matrix of outputs as:

$$
\text{Attention}(Q, K, V) = \text{softmax}(\frac{QK^T}{\sqrt{d_k}})V \tag{1}
$$

The two most commonly used attention functions are additive attention [2], and dot-product (multiplicative) attention. Dot-product attention is identical to our algorithm, except for the scaling factor of $\frac{1}{\sqrt{d_k}}$. Additive attention computes the compatibility function using a feed-forward network with a single hidden layer. While the two are similar in theoretical complexity, dot-product attention is much faster and more space-efficient in practice, since it can be implemented using highly optimized matrix multiplication code.

While for small values of $d_k$ the two mechanisms perform similarly, additive attention outperforms dot product attention without scaling for larger values of $d_k$ [3]. We suspect that for large values of $d_k$, the dot products grow large in magnitude, pushing the softmax function into regions where it has extremely small gradients $^{4}$. To counteract this effect, we scale the dot products by $\frac{1}{\sqrt{d_k}}$.

---

我々の注目すべき点は「スケーリングされたドット積アテンション」（図2）です。入力は、次元 $d_k$ のクエリとキー、および次元 $d_v$ の値から構成されます。クエリとすべてのキーのドット積を計算し、それぞれを $\sqrt{d_k}$ で除算してから、ソフトマックス関数を適用して値に対する重みを取得します。

実際には、一連のクエリに対して同時にアテンション関数を計算し、それらを行列 $Q$ にまとめてパックします。キーと値もまた、行列 $K$ および $V$ にそれぞれまとめられます。出力の行列は以下のように計算されます：

$$
\text{Attention}(Q, K, V) = \text{softmax}(\frac{QK^T}{\sqrt{d_k}})V \tag{1}
$$

最も一般的に使用されるアテンション関数は、加法的アテンション [2] とドット積（乗法的）アテンションです。ドット積アテンションは、$\frac{1}{\sqrt{d_k}}$ というスケールファクターを除いて、我々のアルゴリズムと同じものです。加法的アテンションは、単一の隠れ層を持つフィードフォワードネットワークを使用して適合性関数を計算します。理論的な複雑さにおいて両者は似ていますが、ドット積アテンションは高度に最適化された行列乗算コードで実装できるため、実際にはるかに高速かつスペース効率的です。

$d_k$ が小さい場合、両方のメカニズムは同様の性能を示しますが、$d_k$ が大きい場合、加法的アテンションはスケールなしのドット積アテンションよりも優れた性能を発揮します [3]。我々は、$d_k$ が大きい場合、ドット積の絶対値が大きくなり、ソフトマックス関数が極めて小さな勾配を持つ領域に押しやられるためだと推測しています $^{4}$。この効果を相殺するために、ドット積を $\frac{1}{\sqrt{d_k}}$ によってスケーリングします。

---

## 3.2.2 Multi-Head Attention

---

## 3.2.2 マルチヘッドアテンション

---

Instead of performing a single attention function with $d_{\text{model}}$-dimensional keys, values and queries, we found it beneficial to linearly project the queries, keys and values $h$ times with different, learned linear projections to $d_k$, $d_k$ and $d_v$ dimensions, respectively. On each of these projected versions of queries, keys and values we then perform the attention function in parallel, yielding $d_v$-dimensional

---

$d_{\text{model}}$ 次元のキー、バリュー、クエリに対して単一の注意関数を実行するのではなく、異なる学習済みの線形射影を用いてクエリ、キー、バリューをそれぞれ $d_k$、$d_k$、$d_v$ 次元に $h$ 回線形射影することが有益であることがわかりました。その後、これらの射影されたクエリ、キー、バリューの各バージョンに対して注意関数を並列で実行し、$d_v$ 次元の結果を得ます。

---

$^4$To illustrate why the dot products get large, assume that the components of $q$ and $k$ are independent random variables with mean 0 and variance 1. Then their dot product, $q \cdot k = \sum_{i=1}^{d_k} q_i k_i$, has mean 0 and variance $d_k$.

---

$^4$ドット積が大きくなる理由を説明するために、$q$ と $k$ の成分が平均 0、分散 1 の独立な確率変数であると仮定する。このとき、それらのドット積 $q \cdot k = \sum_{i=1}^{d_k} q_i k_i$ は平均 0、分散 $d_k$ を持つ。

---

output values. These are concatenated and once again projected, resulting in the final values, as depicted in Figure 2.

Multi-head attention allows the model to jointly attend to information from different representation subspaces at different positions. With a single attention head, averaging inhibits this.

$$
\begin{align}
\text{MultiHead}(Q, K, V) &= \text{Concat}(\text{head}_1, ..., \text{head}_{\text{h}})W^O \tag{*}\\
\text{where } \text{head}_{\text{i}} &= \text{Attention}(QW_i^Q, KW_i^K, VW_i^V) \tag{*}
\end{align}
$$

Where the projections are parameter matrices $W_i^Q \in \mathbb{R}^{d_{\text{model}} \times d_k}$, $W_i^K \in \mathbb{R}^{d_{\text{model}} \times d_k}$, $W_i^V \in \mathbb{R}^{d_{\text{model}} \times d_v}$ and $W^O \in \mathbb{R}^{hd_v \times d_{\text{model}}}$.

In this work we employ $h = 8$ parallel attention layers, or heads. For each of these we use $d_k = d_v = d_{\text{model}}/h = 64$. Due to the reduced dimension of each head, the total computational cost is similar to that of single-head attention with full dimensionality.

---

出力値。これらは連結され、再度射影されることで最終的な値が得られ、図2に示されています。

マルチヘッドアテンションにより、モデルは異なる位置において、異なる表現部分空間からの情報を同時に注意深く参照することができます。単一の注意ヘッドでは、平均化がこの機能を阻害します。

$$
\begin{align}
\text{MultiHead}(Q, K, V) &= \text{Concat}(\text{head}_1, ..., \text{head}_{\text{h}})W^O \tag{*}\\
\text{where } \text{head}_{\text{i}} &= \text{Attention}(QW_i^Q, KW_i^K, VW_i^V) \tag{*}
\end{align}
$$

ここで、射影行列は $W_i^Q \in \mathbb{R}^{d_{\text{model}} \times d_k}$、$W_i^K \in \mathbb{R}^{d_{\text{model}} \times d_k}$、$W_i^V \in \mathbb{R}^{d_{\text{model}} \times d_v}$ および $W^O \in \mathbb{R}^{hd_v \times d_{\text{model}}}$ です。

本研究では、並列の注意層（またはヘッド）として $h = 8$ を採用しています。それぞれについて、$d_k = d_v = d_{\text{model}}/h = 64$ を使用します。各ヘッドの次元が削減されているため、総計算コストはフル次元の単一ヘッドアテンションと同程度になります。

---

## 3.2.3 Applications of Attention in our Model

---

## 3.2.3 当モデルにおけるアテンションの応用

---

The Transformer uses multi-head attention in three different ways:

*   In "encoder-decoder attention" layers, the queries come from the previous decoder layer, and the memory keys and values come from the output of the encoder. This allows every position in the decoder to attend over all positions in the input sequence. This mimics the typical encoder-decoder attention mechanisms in sequence-to-sequence models such as [38, 2, 9].
*   The encoder contains self-attention layers. In a self-attention layer all of the keys, values and queries come from the same place, in this case, the output of the previous layer in the encoder. Each position in the encoder can attend to all positions in the previous layer of the encoder.
*   Similarly, self-attention layers in the decoder allow each position in the decoder to attend to all positions in the decoder up to and including that position. We need to prevent leftward information flow in the decoder to preserve the auto-regressive property. We implement this inside of scaled dot-product attention by masking out (setting to $-\infty$) all values in the input of the softmax which correspond to illegal connections. See Figure 2.

---

Transformerは、3つの異なる方法でマルチヘッドアテンションを使用します：

*   「エンコーダ・デコーダ・アテンション」層では、クエリは前のデコーダ層から来ており、キーとバリューはエンコーダの出力から来ます。これにより、デコーダ内のすべての位置が入力シーケンス内のすべての位置にアテンションを向けることができます。これは、[38, 2, 9]などのシーケンスツーシーケンスモデルにおける典型的なエンコーダ・デコーダ・アテンション機構を模倣しています。
*   エンコーダには自己アテンション層が含まれています。自己アテンション層では、すべてのキー、バリュー、およびクエリが同じ場所、つまりエンコーダ内の前の層の出力から来ます。エンコーダ内の各位置は、エンコーダの前の層内のすべての位置にアテンションを向けることができます。
*   同様に、デコーダ内の自己アテンション層は、デコーダ内の各位置がその位置を含むまでのデコーダ内のすべての位置にアテンションを向けることを可能にします。オートリーグレッシブな性質を保つために、デコーダ内での左方向の情報フローを防ぐ必要があります。これは、softmaxの入力に対応する無効な接続に対するすべての値をマスク（$-\infty$ に設定）することで、スケーリングドット積アテンション内で実装されます。図2を参照してください。

---

## 3.3 Position-wise Feed-Forward Networks

---

## 3.3 位置ごとのフィードフォワードネットワーク

---

In addition to attention sub-layers, each of the layers in our encoder and decoder contains a fully connected feed-forward network, which is applied to each position separately and identically. This consists of two linear transformations with a ReLU activation in between.

$$
\mathrm{FFN}(x) = \max(0, xW_1 + b_1)W_2 + b_2 \tag{2}
$$

While the linear transformations are the same across different positions, they use different parameters from layer to layer. Another way of describing this is as two convolutions with kernel size 1. The dimensionality of input and output is $d_{\mathrm{model}} = 512$, and the inner-layer has dimensionality $d_{ff} = 2048$.

---

アテンションのサブレイヤーに加えて、エンコーダとデコーダの各レイヤーには、位置ごとに個別かつ同一に適用される全結合フィードフォワードネットワークが含まれています。これは、ReLU活性化関数を間に挟んだ2つの線形変換から構成されます。

$$
\mathrm{FFN}(x) = \max(0, xW_1 + b_1)W_2 + b_2 \tag{2}
$$

線形変換は異なる位置間で同じですが、レイヤー間では異なるパラメータを使用します。これを別の方法で記述すると、カーネルサイズが1の2つの畳み込みとなります。入出力の次元は $d_{\mathrm{model}} = 512$ で、内部層の次元は $d_{ff} = 2048$ です。

---

## 3.4 Embeddings and Softmax

---

## 3.4 エンベディングとソフトマックス

---

Similarly to other sequence transduction models, we use learned embeddings to convert the input tokens and output tokens to vectors of dimension $d_{\text{model}}$. We also use the usual learned linear transformation and softmax function to convert the decoder output to predicted next-token probabilities. In our model, we share the same weight matrix between the two embedding layers and the pre-softmax linear transformation, similar to [30]. In the embedding layers, we multiply those weights by $\sqrt{d_{\text{model}}}$.

---

他のシーケンス変換モデルと同様に、入力トークンと出力トークンを次元 $d_{\text{model}}$ のベクトルに変換するために学習済み埋め込みを使用します。また、デコーダの出力を次のトークンの確率予測に変換するために、通常の学習済み線形変換とソフトマックス関数も使用します。本モデルでは、2つの埋め込み層とソフトマックス前の線形変換の間で重み行列を共有しており、これは[30]と同様です。埋め込み層では、これらの重みに $\sqrt{d_{\text{model}}}$ を乗じます。

---

Table 1: Maximum path lengths, per-layer complexity and minimum number of sequential operations for different layer types. $n$ is the sequence length, $d$ is the representation dimension, $k$ is the kernel size of convolutions and $r$ the size of the neighborhood in restricted self-attention.

---

表1: 異なるレイヤータイプにおける最大パス長、層ごとの計算量、および最小の逐次演算数。$n$ はシーケンス長、$d$ は表現次元、$k$ は畳み込みのカーネルサイズ、$r$ は制限付き自己注意における近傍領域のサイズです。

---

| Layer Type | Complexity per Layer | Sequential Operations | Maximum Path Length |
| :--- | :---: | :---: | :---: |
| Self-Attention | $O(n^2 \cdot d)$ | $O(1)$ | $O(1)$ |
| Recurrent | $O(n \cdot d^2)$ | $O(n)$ | $O(n)$ |
| Convolutional | $O(k \cdot n \cdot d^2)$ | $O(1)$ | $O(log_k(n))$ |
| Self-Attention (restricted) | $O(r \cdot n \cdot d)$ | $O(1)$ | $O(n/r)$ |

---

| レイヤータイプ | 1レイヤーあたりの計算量 | 逐次演算数 | 最大パス長 |
| :--- | :---: | :---: | :---: |
| セルフアテンション | $O(n^2 \cdot d)$ | $O(1)$ | $O(1)$ |
| リカレント | $O(n \cdot d^2)$ | $O(n)$ | $O(n)$ |
| コンボリューション | $O(k \cdot n \cdot d^2)$ | $O(1)$ | $O(\log_k(n))$ |
| セルフアテンション（制限あり） | $O(r \cdot n \cdot d)$ | $O(1)$ | $O(n/r)$ |

---

## 3.5 Positional Encoding

---

## 3.5 位置エンコーディング

---

Since our model contains no recurrence and no convolution, in order for the model to make use of the order of the sequence, we must inject some information about the relative or absolute position of the tokens in the sequence. To this end, we add "positional encodings" to the input embeddings at the bottoms of the encoder and decoder stacks. The positional encodings have the same dimension $d_{\text{model}}$ as the embeddings, so that the two can be summed. There are many choices of positional encodings, learned and fixed [9].

In this work, we use sine and cosine functions of different frequencies:

$$
\begin{align}
&PE_{(pos,2i)} = sin(pos/10000^{2i/d_{\text{model}}}) \\
&PE_{(pos,2i+1)} = cos(pos/10000^{2i/d_{\text{model}}})
\end{align}
$$

where $pos$ is the position and $i$ is the dimension. That is, each dimension of the positional encoding corresponds to a sinusoid. The wavelengths form a geometric progression from $2\pi$ to $10000 \cdot 2\pi$. We chose this function because we hypothesized it would allow the model to easily learn to attend by relative positions, since for any fixed offset $k$, $PE_{pos+k}$ can be represented as a linear function of $PE_{pos}$.

We also experimented with using learned positional embeddings [9] instead, and found that the two versions produced nearly identical results (see Table 3 row (E)). We chose the sinusoidal version because it may allow the model to extrapolate to sequence lengths longer than the ones encountered during training.

---

モデルに再帰や畳み込みが含まれていないため、モデルがシーケンスの順序を利用するためには、トークンの相対的または絶対的な位置に関する情報を注入する必要があります。そのために、エンコーダとデコーダのスタックの底部にある入力埋め込みに「位置エンコーディング」を追加します。位置エンコーディングは埋め込みと同じ次元 $d_{\text{model}}$ を持ち、これにより両者を加算することができます。位置エンコーディングには多くの選択肢があり、学習済みまたは固定のものがあります [9]。

本研究では、異なる周波数の正弦関数と余弦関数を使用します：

$$
\begin{align}
&PE_{(pos,2i)} = sin(pos/10000^{2i/d_{\text{model}}}) \\
&PE_{(pos,2i+1)} = cos(pos/10000^{2i/d_{\text{model}}})
\end{align}
$$

ここで、$pos$ は位置、$i$ は次元です。つまり、位置エンコーディングの各次元は正弦波に対応しています。波長は $2\pi$ から $10000 \cdot 2\pi$ まで幾何級数的に変化します。この関数を選択したのは、任意の固定オフセット $k$ に対して $PE_{pos+k}$ が $PE_{pos}$ の線形関数として表現できるため、モデルが相対位置による注意を容易に学習できると仮定したからです。

また、代わりに学習済みの位置埋め込み [9] を使用することも実験しましたが、両方のバージョンはほぼ同じ結果を生むことがわかりました（表3の行(E)参照）。正弦波版を選択したのは、それがモデルにとってトレーニング中に遭遇したよりも長いシーケンス長への外挿を可能にする可能性があるためです。

---

## 4 Why Self-Attention

---

## 4 なぜ自己注意なのか

---

In this section we compare various aspects of self-attention layers to the recurrent and convolutional layers commonly used for mapping one variable-length sequence of symbol representations $(x_1, ..., x_n)$ to another sequence of equal length $(z_1, ..., z_n)$, with $x_i, z_i \in \mathbb{R}^d$, such as a hidden layer in a typical sequence transduction encoder or decoder. Motivating our use of self-attention we consider three desiderata.

One is the total computational complexity per layer. Another is the amount of computation that can be parallelized, as measured by the minimum number of sequential operations required.

The third is the path length between long-range dependencies in the network. Learning long-range dependencies is a key challenge in many sequence transduction tasks. One key factor affecting the ability to learn such dependencies is the length of the paths forward and backward signals have to traverse in the network. The shorter these paths between any combination of positions in the input and output sequences, the easier it is to learn long-range dependencies [12]. Hence we also compare the maximum path length between any two input and output positions in networks composed of the different layer types.

As noted in Table 1, a self-attention layer connects all positions with a constant number of sequentially executed operations, whereas a recurrent layer requires $O(n)$ sequential operations. In terms of computational complexity, self-attention layers are faster than recurrent layers when the sequence

---

このセクションでは、自己注意層を、シンボル表現の可変長シーケンス $(x_1, ..., x_n)$ から同等長の別のシーケンス $(z_1, ..., z_n)$ へマッピングするために一般的に使用される再帰型および畳み込み型レイヤーと比較します。ここで $x_i, z_i \in \mathbb{R}^d$ です。これは典型的なシーケンス変換エンコーダまたはデコーダにおける隠れ層などに相当します。自己注意の使用動機として、3つの望ましい特性を検討します。

1つ目は、レイヤーあたりの総計算複雑性です。2つ目は、並列化可能な計算量であり、これは必要な最小の逐次操作数によって測定されます。

3つ目は、ネットワーク内の遠隔依存関係間のパス長です。遠隔依存関係を学習することは、多くのシーケンス変換タスクにおける重要な課題です。そのような依存関係を学習する能力に影響を与える主要な要因の一つは、順伝播および逆伝播の信号がネットワーク内を通過しなければならないパスの長さです。入力および出力シーケンスの任意の組み合わせの位置間のこれらのパスが短いほど、遠隔依存関係を学習しやすくなります [12]。したがって、異なるレイヤータイプで構成されるネットワークにおいて、任意の入力位置と出力位置間の最大パス長も比較します。

表1に示すように、自己注意層はすべての位置を一定数の逐次実行操作で接続しますが、再帰型レイヤーは $O(n)$ の逐次操作を必要とします。計算複雑性の観点から、シーケンスが

---

length $n$ is smaller than the representation dimensionality $d$, which is most often the case with sentence representations used by state-of-the-art models in machine translations, such as word-piece [38] and byte-pair [31] representations. To improve computational performance for tasks involving very long sequences, self-attention could be restricted to considering only a neighborhood of size $r$ in the input sequence centered around the respective output position. This would increase the maximum path length to $O(n/r)$. We plan to investigate this approach further in future work.

A single convolutional layer with kernel width $k < n$ does not connect all pairs of input and output positions. Doing so requires a stack of $O(n/k)$ convolutional layers in the case of contiguous kernels, or $O(\log_k(n))$ in the case of dilated convolutions [18], increasing the length of the longest paths between any two positions in the network. Convolutional layers are generally more expensive than recurrent layers, by a factor of $k$. Separable convolutions [6], however, decrease the complexity considerably, to $O(k \cdot n \cdot d + n \cdot d^2)$. Even with $k = n$, however, the complexity of a separable convolution is equal to the combination of a self-attention layer and a point-wise feed-forward layer, the approach we take in our model.

As side benefit, self-attention could yield more interpretable models. We inspect attention distributions from our models and present and discuss examples in the appendix. Not only do individual attention heads clearly learn to perform different tasks, many appear to exhibit behavior related to the syntactic and semantic structure of the sentences.

---

長さ $n$ が表現次元数 $d$ より小さい場合がほとんどであり、これは機械翻訳における最先端モデルで用いられる文の表現（例えば word-piece [38] や byte-pair [31] 表現）に最もよく当てはまります。非常に長いシーケンスを扱うタスクにおいて計算性能を向上させるために、自己注意機構では入力行列に対して各出力位置を中心としたサイズ $r$ の近傍のみを対象とすることを制限できます。これにより、最長のパス長は $O(n/r)$ に増加します。このアプローチについては今後の研究でさらに検討する予定です。

カーネル幅 $k < n$ の単一畳み込み層では、すべての入力と出力の位置のペアが接続されません。これを達成するには、連続するカーネルの場合 $O(n/k)$ 層の畳み込み層のスタック、または拡張畳み込み [18] の場合 $O(\log_k(n))$ 層が必要となり、ネットワーク内の任意の2つの位置間の最長パスの長さが伸びます。一般的に、畳み込み層は再帰型層よりも $k$ 倍のコストがかかります。しかし、分離可能畳み込み [6] では複雑さが大幅に減少し、$O(k \cdot n \cdot d + n \cdot d^2)$ となります。ただし $k = n$ でも、分離可能畳み込みの複雑さは、自己注意層とポイントワイズ順伝播層を組み合わせたものと同じになります。これが我々のモデルで採用しているアプローチです。

副次的な利点として、自己注意機構はより解釈可能なモデルをもたらす可能性があります。我々はモデルからの注意分布を検査し、付録で例を示して議論しています。個々の注意ヘッドが明確に異なるタスクを実行するように学習しているだけでなく、多くのヘッドが文の構文および意味構造に関連した振る舞いを示しているようです。

---

## 5 Training

---

## 5 トレーニング

---

This section describes the training regime for our models.

---

このセクションでは、当社のモデルのトレーニング手法について説明します。

---

## 5.1 Training Data and Batching

---

## 5.1 訓練データとバッチ処理

---

We trained on the standard WMT 2014 English-German dataset consisting of about 4.5 million sentence pairs. Sentences were encoded using byte-pair encoding [3], which has a shared source-target vocabulary of about 37000 tokens. For English-French, we used the significantly larger WMT 2014 English-French dataset consisting of 36M sentences and split tokens into a 32000 word-piece vocabulary [38]. Sentence pairs were batched together by approximate sequence length. Each training batch contained a set of sentence pairs containing approximately 25000 source tokens and 25000 target tokens.

---

標準的なWMT 2014英語-ドイツ語データセット（約450万文ペアから構成）で学習を行った。文はバイトペアエンコーディング[3]を用いてエンコードされ、共有されるソース・ターゲットの語彙サイズは約37,000トークンであった。英語-フランス語については、より大規模なWMT 2014英語-フランス語データセット（3600万文から構成）を使用し、トークンを32,000語ピースの語彙に分割した[38]。文ペアは概ねシーケンス長に基づいてバッチ化された。各トレーニングバッチには、約25,000個のソーストークンと25,000個のターゲットトークンを含む文ペアのセットが含まれていた。

---

## **5.2 Hardware and Schedule**

---

## **5.2 ハードウェアとスケジュール**

---

We trained our models on one machine with 8 NVIDIA P100 GPUs. For our base models using the hyperparameters described throughout the paper, each training step took about 0.4 seconds. We trained the base models for a total of 100,000 steps or 12 hours. For our big models,(described on the bottom line of table 3), step time was 1.0 seconds. The big models were trained for 300,000 steps (3.5 days).

---

8基のNVIDIA P100 GPUを搭載した1台のマシンでモデルを学習しました。論文全体で説明されているハイパーパラメータを使用するベースモデルの場合、各トレーニングステップは約0.4秒かかりました。ベースモデルは合計100,000ステップ（12時間）学習しました。ビッグモデルについては（表3の最下行に記述）、ステップ時間は1.0秒でした。ビッグモデルは300,000ステップ（3.5日）学習しました。

---

## 5.3 Optimizer

---

## 5.3 最適化アルゴリズム

---

We used the Adam optimizer [20] with $\beta_1 = 0.9$, $\beta_2 = 0.98$ and $\epsilon = 10^{-9}$. We varied the learning rate over the course of training, according to the formula:

$$
lrate = d_{\text{model}}^{-0.5} \cdot \min(step\_num^{-0.5}, step\_num \cdot warmup\_steps^{-1.5}) \tag{3}
$$

This corresponds to increasing the learning rate linearly for the first $warmup\_steps$ training steps, and decreasing it thereafter proportionally to the inverse square root of the step number. We used $warmup\_steps = 4000$.

---

Adam最適化器[20]を使用し、$\beta_1 = 0.9$、$\beta_2 = 0.98$、$\epsilon = 10^{-9}$ とした。学習率は訓練の進行に応じて次の式に従って変化させた：

$$
lrate = d_{\text{model}}^{-0.5} \cdot \min(step\_num^{-0.5}, step\_num \cdot warmup\_steps^{-1.5}) \tag{3}
$$

これは、最初の $warmup\_steps$ 回の訓練ステップでは学習率を線形に増加させ、その後ステップ数の逆平方根に比例して減少させることを意味する。ここでは $warmup\_steps = 4000$ を用いた。

---

## 5.4 Regularization

---

## 5.4 正則化

---

We employ three types of regularization during training:

---

トレーニング中は、3種類の正則化を採用しています。

---

Table 2: The Transformer achieves better BLEU scores than previous state-of-the-art models on the English-to-German and English-to-French newstest2014 tests at a fraction of the training cost.

---

表2: Transformerは、英語からドイツ語および英語からフランス語へのnewstest2014テストにおいて、従来最高のモデルよりも高いBLEUスコアを達成し、かつトレーニングコストの大幅な削減を実現しています。

---

| Model | EN-DE | EN-FR | EN-DE | EN-FR |
| :--- | :---: | :---: | :---: | :---: |
| ByteNet [18] | 23.75 | | | |
| Deep-Att + PosUnk [39] | | 39.2 | | $1.0 \cdot 10^{20}$ |
| GNMT + RL [38] | 24.6 | 39.92 | $2.3 \cdot 10^{19}$ | $1.4 \cdot 10^{20}$ |
| ConvS2S [9] | 25.16 | 40.46 | $9.6 \cdot 10^{18}$ | $1.5 \cdot 10^{20}$ |
| MoE [32] | 26.03 | 40.56 | $2.0 \cdot 10^{19}$ | $1.2 \cdot 10^{20}$ |
| Deep-Att + PosUnk Ensemble [39] | | 40.4 | | $8.0 \cdot 10^{20}$ |
| GNMT + RL Ensemble [38] | 26.30 | 41.16 | $1.8 \cdot 10^{20}$ | $1.1 \cdot 10^{21}$ |
| ConvS2S Ensemble [9] | 26.36 | **41.29** | $7.7 \cdot 10^{19}$ | $1.2 \cdot 10^{21}$ |
| Transformer (base model) | 27.3 | 38.1 | $\mathbf{3.3 \cdot 10^{18}}$ | |
| Transformer (big) | **28.4** | **41.8** | $2.3 \cdot 10^{19}$ | |

**Residual Dropout** We apply dropout [33] to the output of each sub-layer, before it is added to the sub-layer input and normalized. In addition, we apply dropout to the sums of the embeddings and the positional encodings in both the encoder and decoder stacks. For the base model, we use a rate of $P_{drop} = 0.1$.

---

**残差ドロップアウト** 各サブレイヤーの出力に対して、それをサブレイヤーの入力に加算し正規化する前にドロップアウト [33] を適用します。さらに、エンコーダとデコーダのスタックにおいて、埋め込みと位置エンコーディングの和に対してもドロップアウトを適用します。ベースモデルでは、ドロップアウト率を $P_{drop} = 0.1$ とします。

---

**Label Smoothing** During training, we employed label smoothing of value $\epsilon_{ls} = 0.1$ [36]. This hurts perplexity, as the model learns to be more unsure, but improves accuracy and BLEU score.

---

**ラベルスムージング** 訓練中は、値 $\epsilon_{ls} = 0.1$ のラベルスムージング [36] を採用しました。これによりモデルがより不確実な状態を学習するためパープレキシティは悪化しますが、精度とBLEUスコアは向上します。

---

## 6 Results

---

## 6 結果

---

## 6.1 Machine Translation

---

## 6.1 機械翻訳

---

On the WMT 2014 English-to-German translation task, the big transformer model (Transformer (big) in Table 2) outperforms the best previously reported models (including ensembles) by more than 2.0 BLEU, establishing a new state-of-the-art BLEU score of 28.4. The configuration of this model is listed in the bottom line of Table 3. Training took 3.5 days on 8 P100 GPUs. Even our base model surpasses all previously published models and ensembles, at a fraction of the training cost of any of the competitive models.

On the WMT 2014 English-to-French translation task, our big model achieves a BLEU score of 41.0, outperforming all of the previously published single models, at less than $1/4$ the training cost of the previous state-of-the-art model. The Transformer (big) model trained for English-to-French used dropout rate $P_{drop} = 0.1$, instead of 0.3.

For the base models, we used a single model obtained by averaging the last 5 checkpoints, which were written at 10-minute intervals. For the big models, we averaged the last 20 checkpoints. We used beam search with a beam size of 4 and length penalty $\alpha = 0.6$ [38]. These hyperparameters were chosen after experimentation on the development set. We set the maximum output length during inference to input length + 50, but terminate early when possible [38].

Table 2 summarizes our results and compares our translation quality and training costs to other model architectures from the literature. We estimate the number of floating point operations used to train a model by multiplying the training time, the number of GPUs used, and an estimate of the sustained single-precision floating-point capacity of each GPU $^5$.

---

WMT 2014の英語からドイツ語への翻訳タスクにおいて、大規模Transformerモデル（表2の「Transformer (big)」）は、これまでに報告された最良のモデル（アンサンブルを含む）を2.0 BLEU以上上回り、28.4という新たなstate-of-the-artなBLEUスコアを樹立しました。このモデルの構成は表3の最終行に示されています。トレーニングには8つのP100 GPUで3.5日かかりました。さらに、ベースモデルでさえも、競争力のあるモデルのいずれかのトレーニングコストの僅かな割合で、これまでに発表されたすべてのモデルやアンサンブルを上回っています。

WMT 2014の英語からフランス語への翻訳タスクでは、当社の大規模モデルはBLEUスコア41.0を達成し、これまでに発表された単一モデルすべてを上回り、かつ以前のstate-of-the-artモデルのトレーニングコストの4分の1未満で実現しました。英語からフランス語への翻訳用にトレーニングされたTransformer (big)モデルでは、ドロップアウト率 $P_{drop} = 0.1$ が使用され、0.3ではありませんでした。

ベースモデルについては、10分間隔で保存された最後の5つのチェックポイントを平均化した単一モデルを使用しました。大規模モデルについては、最後の20個のチェックポイントを平均化しました。ビームサーチにはビームサイズ4と長さペナルティ $\alpha = 0.6$ [38] を使用しました。これらのハイパーパラメータは開発セットでの実験を経て選択されました。推論時の最大出力長は入力長 + 50 に設定しましたが、可能な場合は早期終了を行いました [38]。

表2は我々の結果を要約し、他の文献からのモデルアーキテクチャと比較して、翻訳品質とトレーニングコストを示しています。モデルのトレーニングに使用された浮動小数点演算数の数は、トレーニング時間、使用されたGPUの数、および各GPUの持続的な単精度浮動小数点容量の見積もり値 ${}^5$ を乗じることで推定されます。

---

## 6.2 Model Variations

---

## 6.2 モデルのバリエーション

---

To evaluate the importance of different components of the Transformer, we varied our base model in different ways, measuring the change in performance on English-to-German translation on the

---

Transformerの異なるコンポーネントの重要性を評価するために、ベースモデルを様々な方法で変更し、英語からドイツ語への翻訳におけるパフォーマンスの変化を測定しました。

---

$^5$We used values of 2.8, 3.7, 6.0 and 9.5 TFLOPS for K80, K40, M40 and P100, respectively.

---

$^5$K80、K40、M40、P100 に対してそれぞれ 2.8、3.7、6.0、9.5 TFLOPS の値を使用しました。

---

Table 3: Variations on the Transformer architecture. Unlisted values are identical to those of the base model. All metrics are on the English-to-German translation development set, newstest2013. Listed perplexities are per-wordpiece, according to our byte-pair encoding, and should not be compared to per-word perplexities.

---

表3: Transformerアーキテクチャのバリエーション。リストされていない値は基本モデルと同じです。すべての指標は、English-to-German翻訳の開発セットnewstest2013に基づいています。記載されているパープレキシティは、バイトペアエンコーディングに基づく単語片ごとの値であり、単語ごとのパープレキシティとは比較できません。

---

| | $N$ | $d_{\text{model}}$ | $d_{\text{ff}}$ | $h$ | $d_k$ | $d_v$ | $P_{drop}$ | $\epsilon_{ls}$ | train steps | PPL (dev) | BLEU (dev) | params $\times 10^6$ |
| :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: |
| base | 6 | 512 | 2048 | 8 | 64 | 64 | 0.1 | 0.1 | 100K | 4.92 | 25.8 | 65 |
| (A) | | | | 1 | 512 | 512 | | | | 5.29 | 24.9 | |
| | | | | 4 | 128 | 128 | | | | 5.00 | 25.5 | |
| | | | | 16 | 32 | 32 | | | | 4.91 | 25.8 | |
| | | | | 32 | 16 | 16 | | | | 5.01 | 25.4 | |
| (B) | | | | | 16 | | | | | 5.16 | 25.1 | 58 |
| | | | | | 32 | | | | | 5.01 | 25.4 | 60 |
| (C) | 2 | | | | | | | | | 6.11 | 23.7 | 36 |
| | 4 | | | | | | | | | 5.19 | 25.3 | 50 |
| | 8 | | | | | | | | | 4.88 | 25.5 | 80 |
| | | 256 | | | 32 | 32 | | | | 5.75 | 24.5 | 28 |
| | | 1024 | | | 128 | 128 | | | | 4.66 | 26.0 | 168 |
| | | | 1024 | | | | | | | 5.12 | 25.4 | 53 |
| | | | 4096 | | | | | | | 4.75 | 26.2 | 90 |
| (D) | | | | | | | 0.0 | | | 5.77 | 24.6 | |
| | | | | | | | 0.2 | | | 4.95 | 25.5 | |
| | | | | | | | | 0.0 | | 4.67 | 25.3 | |
| | | | | | | | | 0.2 | | 5.47 | 25.7 | |
| (E) | | positional embedding instead of sinusoids | | | | | | | | 4.92 | 25.7 | |
| big | 6 | 1024 | 4096 | 16 | | | 0.3 | | 300K | **4.33** | **26.4** | 213 |

development set, newstest2013. We used beam search as described in the previous section, but no checkpoint averaging. We present these results in Table 3.

In Table 3 rows (A), we vary the number of attention heads and the attention key and value dimensions, keeping the amount of computation constant, as described in Section 3.2.2. While single-head attention is 0.9 BLEU worse than the best setting, quality also drops off with too many heads.

In Table 3 rows (B), we observe that reducing the attention key size $d_k$ hurts model quality. This suggests that determining compatibility is not easy and that a more sophisticated compatibility function than dot product may be beneficial. We further observe in rows (C) and (D) that, as expected, bigger models are better, and dropout is very helpful in avoiding over-fitting. In row (E) we replace our sinusoidal positional encoding with learned positional embeddings [9], and observe nearly identical results to the base model.

---

開発セットは newstest2013 です。前のセクションで説明したビームサーチを使用しましたが、チェックポイントの平均化は行いませんでした。これらの結果を Table 3 に示します。

Table 3 の行 (A) では、Section 3.2.2 で述べたように、計算量を一定に保ちつつ、アテンションヘッドの数とアテンションのキーおよびバリューの次元数を変化させました。シングルヘッドのアテンションは最良の設定と比較して BLEU スコアが 0.9 低くなりますが、ヘッドが多すぎると品質も低下することがわかりました。

Table 3 の行 (B) では、アテンションのキーサイズ $d_k$ を小さくするとモデルの品質が低下することが観察されました。これは、適合性の判定が容易ではなく、ドット積よりも洗練された適合性関数が有益である可能性を示唆しています。さらに、行 (C) および (D) では、予想どおり、より大きなモデルの方が性能が高く、ドロップアウトが過学習の回避に非常に有効であることが確認できました。行 (E) では、正弦波位置エンコーディングを学習可能な位置埋め込み [9] に置き換えましたが、ベースラインモデルとほぼ同等の結果を観察しました。

---

## 6.3 English Constituency Parsing

---

## 6.3 英語の統語木解析

---

To evaluate if the Transformer can generalize to other tasks we performed experiments on English constituency parsing. This task presents specific challenges: the output is subject to strong structural constraints and is significantly longer than the input. Furthermore, RNN sequence-to-sequence models have not been able to attain state-of-the-art results in small-data regimes [37].

We trained a 4-layer transformer with $d_{model} = 1024$ on the Wall Street Journal (WSJ) portion of the Penn Treebank [25], about 40K training sentences. We also trained it in a semi-supervised setting, using the larger high-confidence and BerkleyParser corpora from with approximately 17M sentences [37]. We used a vocabulary of 16K tokens for the WSJ only setting and a vocabulary of 32K tokens for the semi-supervised setting.

We performed only a small number of experiments to select the dropout, both attention and residual (section 5.4), learning rates and beam size on the Section 22 development set, all other parameters remained unchanged from the English-to-German base translation model. During inference, we

---

Transformerが他のタスクに一般化できるかを評価するため、英語の統語木解析（constituency parsing）において実験を行った。このタスクには特有の課題がある：出力は強い構造的制約を受け、入力よりも大幅に長くなる。さらに、RNNのシーケンスツーシーケンスモデルは、小規模データ環境において最先端の結果を達成できていない[37]。

私たちは、Penn Treebank [25]のWall Street Journal (WSJ) 部分（約4万文のトレーニング文）を用いて、$d_{model} = 1024$ の4層Transformerを訓練した。また、約1,700万文を含むより大規模な高信頼度コーパスとBerkleyParserコーパスを用いた半教師あり設定でも訓練を行った[37]。WSJのみを使用する設定では16Kトークンの語彙を使い、半教師あり設定では32Kトークンの語彙を使用した。

セクション22の開発セットにおいて、ドロップアウト（注意機構および残差接続、セクション5.4参照）、学習率、ビームサイズを選択するために少数の実験を行い、その他のパラメータは英語からドイツ語への基本翻訳モデルから変更しなかった。推論時には、

---

Table 4: The Transformer generalizes well to English constituency parsing (Results are on Section 23 of WSJ)

---

表4: Transformerは英語の統語構造解析に良好な汎化性能を示す（結果はWSJのセクション23に基づく）

---

| Parser | Training | WSJ 23 F1 |
| :---: | :---: | :---: |
| Vinyals & Kaiser el al. (2014) [37] | WSJ only, discriminative | 88.3 |
| Petrov et al. (2006) [29] | WSJ only, discriminative | 90.4 |
| Zhu et al. (2013) [40] | WSJ only, discriminative | 90.4 |
| Dyer et al. (2016) [8] | WSJ only, discriminative | 91.7 |
| Transformer (4 layers) | WSJ only, discriminative | 91.3 |
| Zhu et al. (2013) [40] | semi-supervised | 91.3 |
| Huang & Harper (2009) [14] | semi-supervised | 91.3 |
| McClosky et al. (2006) [26] | semi-supervised | 92.1 |
| Vinyals & Kaiser el al. (2014) [37] | semi-supervised | 92.1 |
| Transformer (4 layers) | semi-supervised | 92.7 |
| Luong et al. (2015) [23] | multi-task | 93.0 |
| Dyer et al. (2016) [8] | generative | 93.3 |

---

| パーサー | 訓練 | WSJ 23 F1 |
| :---: | :---: | :---: |
| Vinyals & Kaiser et al. (2014) [37] | WSJのみ、判別型 | 88.3 |
| Petrov et al. (2006) [29] | WSJのみ、判別型 | 90.4 |
| Zhu et al. (2013) [40] | WSJのみ、判別型 | 90.4 |
| Dyer et al. (2016) [8] | WSJのみ、判別型 | 91.7 |
| Transformer（4層） | WSJのみ、判別型 | 91.3 |
| Zhu et al. (2013) [40] | セミ教師あり | 91.3 |
| Huang & Harper (2009) [14] | セミ教師あり | 91.3 |
| McClosky et al. (2006) [26] | セミ教師あり | 92.1 |
| Vinyals & Kaiser et al. (2014) [37] | セミ教師あり | 92.1 |
| Transformer（4層） | セミ教師あり | 92.7 |
| Luong et al. (2015) [23] | マルチタスク | 93.0 |
| Dyer et al. (2016) [8] | 生成型 | 93.3 |

---

increased the maximum output length to input length + 300. We used a beam size of 21 and $\alpha = 0.3$ for both WSJ only and the semi-supervised setting.

Our results in Table 4 show that despite the lack of task-specific tuning our model performs surprisingly well, yielding better results than all previously reported models with the exception of the Recurrent Neural Network Grammar [8].

In contrast to RNN sequence-to-sequence models [37], the Transformer outperforms the Berkeley-Parser [29] even when training only on the WSJ training set of 40K sentences.

---

最大出力長を入力長 + 300 に増加させた。WSJ のみの設定と半教師あり設定の両方で、ビームサイズを 21、$\alpha = 0.3$ を使用した。

表4の結果は、タスク固有のチューニングが欠如しているにもかかわらず、我々のモデルが驚くべき性能を発揮し、再帰的ニューラルネットワーク文法 [8] を除くこれまで報告されたすべてのモデルよりも優れた結果を生み出していることを示している。

RNN シーケンス・ツー・シーケンスモデル [37] と対照的に、Transformer は WSJ のトレーニングセット（4万文）のみでトレーニングした場合でも、Berkeley-Parser [29] を上回る性能を示した。

---

## 7 Conclusion

---

## 7 結論

---

In this work, we presented the Transformer, the first sequence transduction model based entirely on attention, replacing the recurrent layers most commonly used in encoder-decoder architectures with multi-headed self-attention.

For translation tasks, the Transformer can be trained significantly faster than architectures based on recurrent or convolutional layers. On both WMT 2014 English-to-German and WMT 2014 English-to-French translation tasks, we achieve a new state of the art. In the former task our best model outperforms even all previously reported ensembles.

We are excited about the future of attention-based models and plan to apply them to other tasks. We plan to extend the Transformer to problems involving input and output modalities other than text and to investigate local, restricted attention mechanisms to efficiently handle large inputs and outputs such as images, audio and video. Making generation less sequential is another research goals of ours.

The code we used to train and evaluate our models is available at https://github.com/tensorflow/tensor2tensor.

**Acknowledgements** We are grateful to Nal Kalchbrenner and Stephan Gouws for their fruitful comments, corrections and inspiration.

---

本稿では、アテンションのみに基づく最初の系列変換モデルであるTransformerを紹介しました。これは、エンコーダ・デコーダアーキテクチャで一般的に使用される再帰層を、マルチヘッド自己注意機構に置き換えています。

翻訳タスクにおいて、Transformerは再帰層や畳み込み層に基づくアーキテクチャよりも大幅に高速に学習できます。WMT 2014英語→ドイツ語およびWMT 2014英語→フランス語の翻訳タスクにおいて、私たちは新たな最先端結果を達成しました。前者のタスクでは、私たちの最良モデルはこれまで報告されたすべてのアンサンブルを上回りました。

私たちはアテンションベースモデルの将来性に期待しており、他のタスクへの適用を検討しています。Transformerをテキスト以外の入力・出力モダリティを含む問題に拡張し、画像、音声、動画などの大規模な入出力を効率的に処理するために局所的かつ制限付きのアテンションメカニズムを検証する予定です。生成過程をより並列化することも、私たちの研究目標の一つです。

モデルの学習と評価に使用したコードは、https://github.com/tensorflow/tensor2tensor で公開されています。

**謝辞** Nal Kalchbrenner および Stephan Gouws の有益なコメント、訂正、そしてインスピレーションに感謝します。

---

## References

---

## 参考文献

---

[1] Jimmy Lei Ba, Jamie Ryan Kiros, and Geoffrey E Hinton. Layer normalization. *arXiv preprint arXiv:1607.06450*, 2016.

[2] Dzmitry Bahdanau, Kyunghyun Cho, and Yoshua Bengio. Neural machine translation by jointly learning to align and translate. *CoRR*, abs/1409.0473, 2014.

[3] Denny Britz, Anna Goldie, Minh-Thang Luong, and Quoc V. Le. Massive exploration of neural machine translation architectures. *CoRR*, abs/1703.03906, 2017.

[4] Jianpeng Cheng, Li Dong, and Mirella Lapata. Long short-term memory-networks for machine reading. *arXiv preprint arXiv:1601.06733*, 2016.

[5] Kyunghyun Cho, Bart van Merrienboer, Caglar Gulcehre, Fethi Bougares, Holger Schwenk, and Yoshua Bengio. Learning phrase representations using rnn encoder-decoder for statistical machine translation. *CoRR*, abs/1406.1078, 2014.

[6] Francois Chollet. Xception: Deep learning with depthwise separable convolutions. *arXiv preprint arXiv:1610.02357*, 2016.

[7] Junyoung Chung, Çağlar Gülçehre, Kyunghyun Cho, and Yoshua Bengio. Empirical evaluation of gated recurrent neural networks on sequence modeling. *CoRR*, abs/1412.3555, 2014.

[8] Chris Dyer, Adhiguna Kuncoro, Miguel Ballesteros, and Noah A. Smith. Recurrent neural network grammars. In *Proc. of NAACL*, 2016.

[9] Jonas Gehring, Michael Auli, David Grangier, Denis Yarats, and Yann N. Dauphin. Convolutional sequence to sequence learning. *arXiv preprint arXiv:1705.03122v2*, 2017.

[10] Alex Graves. Generating sequences with recurrent neural networks. *arXiv preprint arXiv:1308.0850*, 2013.

[11] Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In *Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition*, pages 770–778, 2016.

[12] Sepp Hochreiter, Yoshua Bengio, Paolo Frasconi, and Jürgen Schmidhuber. Gradient flow in recurrent nets: the difficulty of learning long-term dependencies, 2001.

[13] Sepp Hochreiter and Jürgen Schmidhuber. Long short-term memory. *Neural computation*, 9(8):1735–1780, 1997.

[14] Zhongqiang Huang and Mary Harper. Self-training PCFG grammars with latent annotations across languages. In *Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing*, pages 832–841. ACL, August 2009.

[15] Rafal Jozefowicz, Oriol Vinyals, Mike Schuster, Noam Shazeer, and Yonghui Wu. Exploring the limits of language modeling. *arXiv preprint arXiv:1602.02410*, 2016.

[16] Łukasz Kaiser and Samy Bengio. Can active memory replace attention? In *Advances in Neural Information Processing Systems, (NIPS)*, 2016.

[17] Łukasz Kaiser and Ilya Sutskever. Neural GPUs learn algorithms. In *International Conference on Learning Representations (ICLR)*, 2016.

[18] Nal Kalchbrenner, Lasse Espeholt, Karen Simonyan, Aaron van den Oord, Alex Graves, and Koray Kavukcuoglu. Neural machine translation in linear time. *arXiv preprint arXiv:1610.10099v2*, 2017.

[19] Yoon Kim, Carl Denton, Luong Hoang, and Alexander M. Rush. Structured attention networks. In *International Conference on Learning Representations*, 2017.

[20] Diederik Kingma and Jimmy Ba. Adam: A method for stochastic optimization. In *ICLR*, 2015.

[21] Oleksii Kuchaiev and Boris Ginsburg. Factorization tricks for LSTM networks. *arXiv preprint arXiv:1703.10722*, 2017.

[22] Zhouhan Lin, Minwei Feng, Cicero Nogueira dos Santos, Mo Yu, Bing Xiang, Bowen Zhou, and Yoshua Bengio. A structured self-attentive sentence embedding. *arXiv preprint arXiv:1703.03130*, 2017.

[23] Minh-Thang Luong, Quoc V. Le, Ilya Sutskever, Oriol Vinyals, and Lukasz Kaiser. Multi-task sequence to sequence learning. *arXiv preprint arXiv:1511.06114*, 2015.

[24] Minh-Thang Luong, Hieu Pham, and Christopher D Manning. Effective approaches to attention-based neural machine translation. *arXiv preprint arXiv:1508.04025*, 2015.

---

[5] Kyunghyun Cho, Bart van Merrienboer, Caglar Gulcehre, Fethi Bougares, Holger Schwenk, Yoshua Bengio. 統計的機械翻訳のためのRNNエンコーダ・デコーダを用いたフレーズ表現の学習. *CoRR*, abs/1406.1078, 2014.

[6] Francois Chollet. Xception: ディープラーニングにおける深層分離畳み込み. *arXiv preprint arXiv:1610.02357*, 2016.

[7] Junyoung Chung, Çağlar Gülçehre, Kyunghyun Cho, Yoshua Bengio. 系列モデリングにおけるゲート付き再帰型ニューラルネットワークの実証評価. *CoRR*, abs/1412.3555, 2014.

[8] Chris Dyer, Adhiguna Kuncoro, Miguel Ballesteros, Noah A. Smith. 再帰型ニューラルネットワーク文法. In *Proc. of NAACL*, 2016.

[9] Jonas Gehring, Michael Auli, David Grangier, Denis Yarats, Yann N. Dauphin. 畳み込みシーケンスツーシーケンス学習. *arXiv preprint arXiv:1705.03122v2*, 2017.

[10] Alex Graves. 再帰型ニューラルネットワークによる系列生成. *arXiv preprint arXiv:1308.0850*, 2013.

[11] Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun. 画像認識のための深層残差学習. In *Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition*, pages 770–778, 2016.

[12] Sepp Hochreiter, Yoshua Bengio, Paolo Frasconi, Jürgen Schmidhuber. 再帰型ネットにおける勾配フロー：長期依存性の学習の難しさ, 2001.

[13] Sepp Hochreiter, Jürgen Schmidhuber. Long short-term memory. *Neural computation*, 9(8):1735–1780, 1997.

[14] Zhongqiang Huang, Mary Harper. 言語横断的な潜在注釈を持つ自己トレーニングPCFG文法. In *Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing*, pages 832–841. ACL, August 2009.

[15] Rafal Jozefowicz, Oriol Vinyals, Mike Schuster, Noam Shazeer, Yonghui Wu. 言語モデリングの限界を探る. *arXiv preprint arXiv:1602.02410*, 2016.

[16] Łukasz Kaiser, Samy Bengio. アクティブメモリは注意機構を置き換えることができるか？ In *Advances in Neural Information Processing Systems, (NIPS)*, 2016.

[17] Łukasz Kaiser, Ilya Sutskever. ニューラルGPUはアルゴリズムを学習する. In *International Conference on Learning Representations (ICLR)*, 2016.

[18] Nal Kalchbrenner, Lasse Espeholt, Karen Simonyan, Aaron van den Oord, Alex Graves, Koray Kavukcuoglu. 線形時間でのニューラル機械翻訳. *arXiv preprint arXiv:1610.10099v2*, 2017.

[19] Yoon Kim, Carl Denton, Luong Hoang, Alexander M. Rush. 構造化注意ネットワーク. In *International Conference on Learning Representations*, 2017.

[20] Diederik Kingma, Jimmy Ba. Adam: 確率的最適化のための手法. In *ICLR*, 2015.

[21] Oleksii Kuchaiev, Boris Ginsburg. LSTMネットワークのための因子分解トリック. *arXiv preprint arXiv:1703.10722*, 2017.

[22] Zhouhan Lin, Minwei Feng, Cicero Nogueira dos Santos, Mo Yu, Bing Xiang, Bowen Zhou, Yoshua Bengio. 構造化自己注意文埋め込み. *arXiv preprint arXiv:1703.03130*, 2017.

[23] Minh-Thang Luong, Quoc V. Le, Ilya Sutskever, Oriol Vinyals, Lukasz Kaiser. マルチタスクシーケンスツーシーケンス学習. *arXiv preprint arXiv:1511.06114*, 2015.

[24] Minh-Thang Luong, Hieu Pham, Christopher D Manning. 注意ベースのニューラル機械翻訳への効果的なアプローチ. *arXiv preprint arXiv:1508.04025*, 2015.

---

[25] Mitchell P Marcus, Mary Ann Marcinkiewicz, and Beatrice Santorini. Building a large annotated corpus of english: The penn treebank. *Computational linguistics*, 19(2):313–330, 1993.

[26] David McClosky, Eugene Charniak, and Mark Johnson. Effective self-training for parsing. In *Proceedings of the Human Language Technology Conference of the NAACL, Main Conference*, pages 152–159. ACL, June 2006.

[27] Ankur Parikh, Oscar Täckström, Dipanjan Das, and Jakob Uszkoreit. A decomposable attention model. In *Empirical Methods in Natural Language Processing*, 2016.

[28] Romain Paulus, Caiming Xiong, and Richard Socher. A deep reinforced model for abstractive summarization. *arXiv preprint arXiv:1705.04304*, 2017.

[29] Slav Petrov, Leon Barrett, Romain Thibaux, and Dan Klein. Learning accurate, compact, and interpretable tree annotation. In *Proceedings of the 21st International Conference on Computational Linguistics and 44th Annual Meeting of the ACL*, pages 433–440. ACL, July 2006.

[30] Ofir Press and Lior Wolf. Using the output embedding to improve language models. *arXiv preprint arXiv:1608.05859*, 2016.

[31] Rico Sennrich, Barry Haddow, and Alexandra Birch. Neural machine translation of rare words with subword units. *arXiv preprint arXiv:1508.07909*, 2015.

[32] Noam Shazeer, Azalia Mirhoseini, Krzysztof Maziarz, Andy Davis, Quoc Le, Geoffrey Hinton, and Jeff Dean. Outrageously large neural networks: The sparsely-gated mixture-of-experts layer. *arXiv preprint arXiv:1701.06538*, 2017.

[33] Nitish Srivastava, Geoffrey E Hinton, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdinov. Dropout: a simple way to prevent neural networks from overfitting. *Journal of Machine Learning Research*, 15(1):1929–1958, 2014.

[34] Sainbayar Sukhbaatar, Arthur Szlam, Jason Weston, and Rob Fergus. End-to-end memory networks. In C. Cortes, N. D. Lawrence, D. D. Lee, M. Sugiyama, and R. Garnett, editors, *Advances in Neural Information Processing Systems 28*, pages 2440–2448. Curran Associates, Inc., 2015.

[35] Ilya Sutskever, Oriol Vinyals, and Quoc VV Le. Sequence to sequence learning with neural networks. In *Advances in Neural Information Processing Systems*, pages 3104–3112, 2014.

[36] Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jonathon Shlens, and Zbigniew Wojna. Rethinking the inception architecture for computer vision. *CoRR*, abs/1512.00567, 2015.

[37] Vinyals & Kaiser, Koo, Petrov, Sutskever, and Hinton. Grammar as a foreign language. In *Advances in Neural Information Processing Systems*, 2015.

[38] Yonghui Wu, Mike Schuster, Zhifeng Chen, Quoc V Le, Mohammad Norouzi, Wolfgang Macherey, Maxim Krikun, Yuan Cao, Qin Gao, Klaus Macherey, et al. Google’s neural machine translation system: Bridging the gap between human and machine translation. *arXiv preprint arXiv:1609.08144*, 2016.

[39] Jie Zhou, Ying Cao, Xuguang Wang, Peng Li, and Wei Xu. Deep recurrent models with fast-forward connections for neural machine translation. *CoRR*, abs/1606.04199, 2016.

[40] Muhua Zhu, Yue Zhang, Wenliang Chen, Min Zhang, and Jingbo Zhu. Fast and accurate shift-reduce constituent parsing. In *Proceedings of the 51st Annual Meeting of the ACL (Volume 1: Long Papers)*, pages 434–443. ACL, August 2013.

---

[25] Mitchell P Marcus, Mary Ann Marcinkiewicz, and Beatrice Santorini. 大規模注釈付き英語コーパスの構築：ペンツリーバンク. *計算言語学*, 19(2):313–330, 1993.

[26] David McClosky, Eugene Charniak, and Mark Johnson. 構文解析のための効果的な自己教師あり学習. In *NAACL人間言語技術会議本会議論文集*, pages 152–159. ACL, 2006年6月.

[27] Ankur Parikh, Oscar Täckström, Dipanjan Das, and Jakob Uszkoreit. 分解可能アテンションモデル. In *自然言語処理における実証的アプローチ*, 2016.

[28] Romain Paulus, Caiming Xiong, and Richard Socher. 要約生成のための深層強化学習モデル. *arXivプレプリント arXiv:1705.04304*, 2017.

[29] Slav Petrov, Leon Barrett, Romain Thibaux, and Dan Klein. 正確でコンパクト、かつ解釈可能な木構造注釈の学習. In *第21回国際計算言語学会議および第44回ACL年次会議論文集*, pages 433–440. ACL, 2006年7月.

[30] Ofir Press and Lior Wolf. 出力埋め込みを用いた言語モデルの改善. *arXivプレプリント arXiv:1608.05859*, 2016.

[31] Rico Sennrich, Barry Haddow, and Alexandra Birch. サブワード単位による稀な単語のニューラル機械翻訳. *arXivプレプリント arXiv:1508.07909*, 2015.

[32] Noam Shazeer, Azalia Mirhoseini, Krzysztof Maziarz, Andy Davis, Quoc Le, Geoffrey Hinton, and Jeff Dean. 驚異的に大規模なニューラルネットワーク：スパースにゲートされたエキスパート混合レイヤー. *arXivプレプリント arXiv:1701.06538*, 2017.

[33] Nitish Srivastava, Geoffrey E Hinton, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdinov. Dropout: ニューラルネットワークの過学習を防ぐ簡単な方法. *Journal of Machine Learning Research*, 15(1):1929–1958, 2014.

[34] Sainbayar Sukhbaatar, Arthur Szlam, Jason Weston, and Rob Fergus. エンドツーエンドメモリネットワーク. In C. Cortes, N. D. Lawrence, D. D. Lee, M. Sugiyama, and R. Garnett, editors, *ニューラル情報処理システム進歩 28*, pages 2440–2448. Curran Associates, Inc., 2015.

[35] Ilya Sutskever, Oriol Vinyals, and Quoc VV Le. ニューラルネットワークを用いたシーケンスからシーケンスへの学習. In *ニューラル情報処理システム進歩*, pages 3104–3112, 2014.

[36] Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jonathon Shlens, and Zbigniew Wojna. コンピュータビジョンのためのインセプションアーキテクチャの再考. *CoRR*, abs/1512.00567, 2015.

[37] Vinyals & Kaiser, Koo, Petrov, Sutskever, and Hinton. 外国語としての文法. In *ニューラル情報処理システム進歩*, 2015.

[38] Yonghui Wu, Mike Schuster, Zhifeng Chen, Quoc V Le, Mohammad Norouzi, Wolfgang Macherey, Maxim Krikun, Yuan Cao, Qin Gao, Klaus Macherey, et al. Googleのニューラル機械翻訳システム：人間の翻訳と機械翻訳のギャップを埋める. *arXivプレプリント arXiv:1609.08144*, 2016.

[39] Jie Zhou, Ying Cao, Xuguang Wang, Peng Li, and Wei Xu. フォワード接続を持つ深層再帰モデルによるニューラル機械翻訳. *CoRR*, abs/1606.04199, 2016.

[40] Muhua Zhu, Yue Zhang, Wenliang Chen, Min Zhang, and Jingbo Zhu. 高速かつ正確なシフト・リダクション構成要素構文解析. In *ACL第51回年次会議論文集（巻1：長編論文）*, pages 434–443. ACL, 2013年8月.

---

---

<img src="images/13-1.png" style="zoom:70%; display: block; margin: 0 auto;" />

Figure 3: An example of the attention mechanism following long-distance dependencies in the encoder self-attention in layer 5 of 6. Many of the attention heads attend to a distant dependency of the verb ‘making’, completing the phrase ‘making...more difficult’. Attentions here shown only for the word ‘making’. Different colors represent different heads. Best viewed in color.

---

図3: 6層中の5番目のレイヤーにおけるエンコーダの自己注意機構が長距離依存関係を捉える例。多くの注意ヘッドは動詞「making」の遠い依存関係に注目し、「making...more difficult」というフレーズを補完している。ここでは単語「making」に対する注意のみを示している。異なる色は異なるヘッドを表す。カラーで確認するのが最適。

---

<img src="images/14-0.png" style="zoom:70%; display: block; margin: 0 auto;" />

Figure 4: Two attention heads, also in layer 5 of 6, apparently involved in anaphora resolution. Top: Full attentions for head 5. Bottom: Isolated attentions from just the word 'its' for attention heads 5 and 6. Note that the attentions are very sharp for this word.

---

図4: 6層中の5層にある2つのアテンションヘッドは、アナフォラ解決に関与しているようだ。上段: ヘッド5の完全なアテンション。下段: アテンションヘッド5と6における単語「its」からの孤立したアテンション。この単語に対するアテンションが非常に鋭いことに注目。

---

<img src="images/15-0.png" style="zoom:70%; display: block; margin: 0 auto;" />

Figure 5: Many of the attention heads exhibit behaviour that seems related to the structure of the sentence. We give two such examples above, from two different heads from the encoder self-attention at layer 5 of 6. The heads clearly learned to perform different tasks.

---

図5: 多くのアテンションヘッドは、文の構造に関連しているように見える振る舞いを示しています。上記では、6層中の5層におけるエンコーダの自己注意機構から2つの異なるヘッドの例を2つ示しています。これらのヘッドは明確に異なるタスクを実行することを学習しました。

---
